Signal and Image Processing in Navigational Systems (Electrical Engineering & Applied Signal Processing Series)

Author: Vyacheslav Tuzlukov

76 downloads 771 Views 10MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!

Report copyright / DMCA form

DOWNLOAD PDF

SIGNAL and image processing in navigational systems

Copyright 2005 by CRC Press

THE ELECTRICAL ENGINEERING AND APPLIED SIGNAL PROCESSING SERIES Edited by Alexander Poularikas The Advanced Signal Processing Handbook: Theory and Implementation for Radar, Sonar, and Medical Imaging Real-Time Systems Stergios Stergiopoulos The Transform and Data Compression Handbook K.R. Rao and P.C. Yip Handbook of Multisensor Data Fusion David Hall and James Llinas Handbook of Neural Network Signal Processing Yu Hen Hu and Jenq-Neng Hwang Handbook of Antennas in Wireless Communications Lal Chand Godara Noise Reduction in Speech Applications Gillian M. Davis Signal Processing Noise Vyacheslav P. Tuzlukov Digital Signal Processing with Examples in MATLAB® Samuel Stearns Applications in Time-Frequency Signal Processing Antonia Papandreou-Suppappola The Digital Color Imaging Handbook Gaurav Sharma Pattern Recognition in Speech and Language Processing Wu Chou and Biing-Hwang Juang Propagation Handbook for Wireless Communication System Design Robert K. Crane Nonlinear Signal and Image Processing: Theory, Methods, and Applications Kenneth E. Barner and Gonzalo R. Arce Smart Antennas Lal Chand Godara Mobile Internet: Enabling Technologies and Services Apostolis K. Salkintzis and Alexander Poularikas Soft Computing with MATLAB® Ali Zilouchian Wireless Internet: Technologies and Applications Apostolis K. Salkintzis and Alexander Poularikas Signal and Image Processing in Navigational Systems Vyacheslav P. Tuzlukov Copyright 2005 by CRC Press

SIGNAL and image processing in navigational systems Vyacheslav P. Tuzlukov

CRC PR E S S Boca Raton London New York Washington, D.C. Copyright 2005 by CRC Press

1598_C00.fm Page iv Friday, October 8, 2004 2:00 PM

Library of Congress Cataloging-in-Publication Data Tuzlukov, V. P. (Vyacheslav Petrovich) Signal and image processing in navigational systems / Vyacheslav Tuzlukov. p. cm. -- (The electrical engineering and applied signal processing series) Includes bibliographical references and index. ISBN 0-8493-1598-0 (alk. paper) 1. Electronics in navigation 2. Signal processing. 3. Image processing. I. Title. II. Series. VK560.T88 2004 623.89′3′0285—dc22

2004049669

This book contains information obtained from authentic and highly regarded sources. Reprinted material is quoted with permission, and sources are indicated. A wide variety of references are listed. Reasonable efforts have been made to publish reliable data and information, but the author and the publisher cannot assume responsibility for the validity of all materials or for the consequences of their use. Neither this book nor any part may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying, microfilming, and recording, or by any information storage or retrieval system, without prior permission in writing from the publisher. The consent of CRC Press does not extend to copying for general distribution, for promotion, for creating new works, or for resale. Specific permission must be obtained in writing from CRC Press for such copying. Direct all inquiries to CRC Press, 2000 N.W. Corporate Blvd., Boca Raton, Florida 33431. Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are used only for identification and explanation, without intent to infringe.

Visit the CRC Press Web site at www.crcpress.com © 2005 by CRC Press No claim to original U.S. Government works International Standard Book Number 0-8493-1598-0 Library of Congress Card Number 2004049669 Printed in the United States of America 1 2 3 4 5 6 7 8 9 0 Printed on acid-free paper

Copyright 2005 by CRC Press

1598_C00.fm Page v Friday, October 8, 2004 2:00 PM

Preface

Noise immunity is the main problem in navigational systems. At the present time, there are many books and journal articles devoted to signal and image processing in noise in navigational systems, but many important problems remain to be solved. New approaches and study of complex problems allow us not only to summarize investigations but also to derive a better quality of signal and image processing in noise in navigational systems. In the functioning of many navigational systems, reflections from the Earth’s surface and the rough or rippled sea, hydrometeors (storm clouds, rain, shower, snow, etc.), the ionosphere, clouds of artificial scatterers, etc., play a large role. We can observe these reflections in the detection and tracking of low-flying, surface- or sea-surface-moving targets against the background of highly camouflaged reflections from the underlying surface. In these cases, reflections play the role of passive interference, and there is the need to construct specific methods and techniques for increasing the noise immunity of navigational systems. There are many navigational systems in which reflections from the Earth’s or sea’s surface and hydrometeors are the information signal and not an interference. Examples are autonomous navigational systems for aircraft with the Doppler analyzer of velocity and drift angle, analyzer of vertical velocity of take-off and landing, height-finding radar; navigational systems with ground surveillance radar, scatter meters in which reflections from the Earth’s surface and rough and rippled sea are used to obtain the detailed information about surface structure and state; storm-warning radar; and weather radar. This book is devoted to the study of fluctuations of parameters of the target return signals and signal and image processing problems in navigational systems constructed on the basis of the generalized approach to signal and image processing in noise based on a seemingly abstract idea: the introduction of an additional noise source that does not carry any information about the signal with the purpose of improving the qualitative performances of complex navigational systems. Theoretical and experimental study carried out by the author leads to the conclusion that the proposed generalized approach to signal and image processing in noise in navigational systems allows us to formulate a decision-making rule based on the determination of the jointly sufficient statistics of the likelihood function (or functional) mean and variance. The use of classical and modern signal and image processing approaches in navigational systems allows us to define only the sufficient statistic of the likelihood function (or functional) mean.

Copyright 2005 by CRC Press

1598_C00.fm Page vi Friday, October 8, 2004 2:00 PM

The presence of additional information about the statistical characteristics of the likelihood function (or functional) leads to better qualitative performances of signal and image processing in navigational systems compared to the optimal signal and image processing algorithms of classical and modern theories. The generalized approach to signal and image processing in navigational systems allows us to extend the well-known boundaries of the potential noise immunity set up by classical and modern signal and image processing theories. The use of complex navigational systems based on the generalized approach allows us to obtain better detection performances and definition of object coordinates with high accuracy, in particular, in comparison with navigational systems constructed on the basis of optimal and asymptotic optimal signal and image processing algorithms of classical and modern theories. To understand better the fundamental statements and concepts of the generalized approach, the reader is invited to consult my earlier books: Signal Processing in Noise: A New Methodology (IEC, Minsk, 1998), Signal Detection Theory (Springer-Verlag, New York, 2001), and Signal Processing Noise (CRC Press, Boca Raton, FL, 2002). I would like to thank my many colleagues in the field of signal and image processing for very useful discussions about the main results, in particular, Professors V. Ignatov, A. Kolyada, I. Malevich, G. Manshin, B. Levin, D. Johnson, B. Bogner, Yu Sedyshev, J. Schroeder, Yu Shinakov, A. Kara, X.R. Lee, Yong Deak Kim, Won-Sik Yoon, V. Kuzkin, A. Dubey, and O. Drummond. A special word of thanks to Ajou University, Suwon, South Korea, for allowing me to complete this project. A lot of credit also needs to go to Nora Konopka, Jessica Vakili, Gail Renard, and the staff at CRC Press for their encouragement and support of this project. Last, but definitely not least, I would like to thank my family, my lovely wife and sons and my dear mother, for putting up with me during the completion of the manuscript; without their support it would not have been possible! I also wish to express my lifelong, heartfelt gratitude to Peter Tuzlukov, my father and teacher, who introduced me to science. Vyacheslav Tuzlukov

Copyright 2005 by CRC Press

1598_C00.fm Page vii Friday, October 8, 2004 2:00 PM

The Author

Vyacheslav Tuzlukov, Ph.D., is Invited Full Professor at Ajou University, Suwon, South Korea, and Chief Research Fellow at the United Institute on Informatics Problems of the National Academy of Sciences, Belarus. He is also a Full Professor in the Electrical and Computer Engineering Department of the Belarussian State University in Minsk. During 2000 to 2002, Dr. Tuzlukov was a Visiting Full Professor at the University of Aizu, AizuWakamatsu, Japan. He is actively engaged in research on radar, communications, and signal and image processing, and has more than 25 years’ experience in these areas. Dr. Tuzlukov is the author of more than 120 journal articles and conference papers and five books — one in Russian and four in English — on signal processing.

Copyright 2005 by CRC Press

1598_C00.fm Page ix Friday, October 8, 2004 2:00 PM

Introduction

The main functioning principle of any navigational system is based on the comparison of the moving image of the Earth’s surface or the totality of landmarks with the reference image or model image. The moving and model images are formed using various natural and manmade physical fields. As an illustration of these fields, we can use optical, radar, radio heat, electromagnetic, gravitational, and other fields. For example, using an airborne radar, we can obtain the moving radar images of the Earth’s surface that are compared with the predetermined images corresponding to the required airborne flight track. The measure of deviation of the flight track from the predetermined or required airborne flight track is characterized by mutual noncoincidence between the moving image and the model image relative to each other. Coincidence between the moving and model images is used to restore an airborne flight track to the true flight track. This working principle of navigational systems is called the searching-free principle. Another example allows us to consider a navigational system based on the searching principle. Let us assume there are data of model images corresponding to all possible airborne flight tracks. Each model image corresponds to the definite coordinate system. Maximum coincidence between the moving image and a model image allows us to define the true airborne flight track. Comparison between moving and model images is made by the functional that is at its maximum in the coincidence between the moving and model images. The mutual correlation function can be considered as a functional with some limitations. In the coincidence between the moving and model images, the correlation function must be maximum and its derivative must be minimum. Due to the stochastic character of elementary scatterers, the amplitude and phase of the target return signal at the receiver or detector input in navigational systems are random variables. For many reasons, such as scatterer moving under the stimulus of the wind, the radar moving, radar antenna scanning, etc., the target return signal at the receiver or detector input is a stochastic process with fluctuating parameters. Therefore, the target return signal at the receiver or detector input in navigational systems can be defined by the probability distribution density, correlation function, and power spectral density that depend on some parameters of radar and navigational system devices and peculiarities of scatterers: the shape of the directional diagram, orientation of the directional diagram with respect to the velocity vector of the radar moving, localization of scatterers in space, the shape of the searching signal, laws of the radar moving and radar antenna scanning, and the nature and character of scatterer moving. The fluctuations of parameters of the target

Copyright 2005 by CRC Press

1598_C00.fm Page x Friday, October 8, 2004 2:00 PM

return signal at the receiver or detector input are caused, as a rule, by simultaneous stimulus of noise and interference sources. For example, in the use of navigational systems based on aircraft radar, we should take into consideration the radar moving, radar antenna scanning, and direction of the wind. In some cases, we have to add instability in the frequency of the transmitting radar antenna and (or) rotation of the polarization plane of the radar antenna caused by radar antenna scanning and rotation in navigational systems to the noise and interference sources mentioned previously. Furthermore, the various power spectral densities of fluctuations of the target return signal at the receiver or detector input can be formed under the stimulus of the same noise and interference sources in accordance with the specific input stochastic process because the radar moving, radar antenna scanning and rotation, and scatterer moving are different in different cases in practice. Because of this, the correlation functions and power spectral densities of fluctuations of the target return signal at the receiver or detector input in navigational systems are specific characteristics of the input stochastic process and must be defined for specific conditions in practice. At the same time, signal processing in navigational systems depends greatly on the correlation features of the target return signal at the receiver or detector input. In particular, the definition of the shape of the power spectral density of passive interferences has a fundamental significance in solving the problem of optimal signal processing and in defining the effectiveness of signal processing in navigational systems. In the case of autonomous navigational systems in which the target return signal from the underlying surface of the Earth or sea possesses information regarding measured parameters, such as velocity, distance, and direction, it is necessary to take into consideration the probability distribution laws, the effective bandwidth, and shape of the power spectral density of the target return signal at the receiver or detector input because all of these allow us to choose the true signal processing technique and algorithms and to ensure a high accuracy of definition in the measured parameters of the target return signal. To construct navigational systems with high noise immunity, we have to define with high accuracy the power spectral density of the target return signal at the receiver or detector input because the effectiveness of the navigational system functioning depends on the knowledge of, for example, the rate of decrease, the length of remainders, deviation from the axis of symmetry of the power spectral density, etc. The function between the power spectral density and various factors can be complex and not always clear. Therefore, in theoretical investigations, the power spectral density with the Gaussian, resonant, and square waveform shape are used for simplicity and convenience of analysis. The real power spectral densities of the target return signal at the receiver or detector input in navigational systems, taking into consideration specific conditions of their forming in practice, would be defined in a rigorous form. This book summarizes investigations carried out by the author over the last 20 years.

Copyright 2005 by CRC Press

1598_C00.fm Page xi Friday, October 8, 2004 2:00 PM

The book consists of two parts. The first part discusses fluctuations of the target return signal parameters in navigational systems based on the generalized approach to signal and image processing in noise. Discussed results presenting the majority of cases in practice take into consideration almost all possible sources of fluctuations of the target return signal parameters. The second part is concerned with navigational systems based on the generalized approach to space–time signal and image processing. Detailed attention is paid to the employment of optical detectors, optical direction finders, and optical coordinate analyzers constructed on the basis of the generalized approach to signal and image processing in noise. The book comprises 14 chapters. Chapter 1 discusses the problems of definition of the probability distribution density of the target return signal amplitude and phase. Two-dimensional probability distribution density is defined. Based on the two-dimensional probability distribution density, we are able to obtain the following particular cases: the probability distribution density of the amplitude and the probability distribution density of the phase. The parameters of the probability distribution density are defined as a function of the distribution law of amplitudes and phases of elementary signals. Particular cases, namely, the uniform, “triangular,” and Gaussian probability distribution densities of the phase are considered. Chapter 2 deals with the study of the correlation function of the target return signal. Physical sources of fluctuations of parameters are investigated. The space–time correlation function and power spectral density are defined. The correlation function with the searching signal of arbitrary shape, for example, the narrow-band searching signal and pulsed searching signal, is discussed. The correlation function in scanning the three-dimensional (space) target is defined in the cases of the pulsed searching signal and the simple harmonic searching signal. The correlation function in angle scanning the two-dimensional (surface) target is studied in the cases of the pulsed searching signal and the simple harmonic searching signal. The correlation function of the target return signal is defined under vertical scanning of the twodimensional (surface) target. Chapter 3 is concerned with the definition of fluctuations of the target return signal parameters in scanning the three-dimensional (space) target by the moving radar. The cases of slow and rapid fluctuations are considered. Additionally, Doppler fluctuations of the target return signal parameters in navigational systems with the high-deflected antenna are investigated in the cases of arbitrary directional diagrams, Gaussian directional diagrams, and sinc-directional diagrams. Doppler fluctuations with the arbitrarily deflected radar antenna are discussed, and the total power spectral density of the target return signal in the case of the pulsed searching signal is defined. Chapter 4 focuses on the definition of fluctuations of the target return signal parameters in scanning the two-dimensional (surface) target by the moving radar. The continuous nonmodulated searching and pulsed searching signals in the stationary radar are considered as initial premises for the following cases: arbitrary vertical-coverage directional diagrams — Gaussian Copyright 2005 by CRC Press

1598_C00.fm Page xii Friday, October 8, 2004 2:00 PM

pulsed and square waveform searching signals, and Gaussian vertical-coverage directional diagrams — square waveform searching signal. In the case of the pulsed searching signal in the moving radar, the angle correlation function is defined under the following conditions: the arbitrary verticalcoverage directional diagram — Gaussian pulsed searching signal, and the Gaussian vertical-coverage directional diagram — Gaussian pulsed searching signal. The azimuth correlation function in the case of the pulsed searching signal in the moving radar is defined at the high- and low-deflected radar antenna. The total correlation function and power spectral density of fluctuations of the target return signal parameters at the receiver or detector input are defined for the following cases: Gaussian directional diagrams and the Gaussian pulsed searching signal, the square waveform searching signal, and the pulsed searching signal with short duration. The minimum radar range is defined under the conditions described earlier. The vertical scanning of the two-dimensional (surface) target is investigated. Examples of determination of the power spectral density of the target return signal are presented. Chapter 5 explores problems of definition of fluctuations of the target return signal parameters caused by radar antenna scanning. The correlation functions in space and surface scanning are defined. The general definition of the power spectral density is discussed. The line radar antenna scanning is investigated for the following cases: one-line T-scanning, multiple-line T-scanning, line segment scanning, and line T-scanning for various directional diagrams in the transmitter–receiver block of the navigational system. Conical antenna scanning is studied in the cases of three-dimensional (space) and two-dimensional (surface) targets. Moreover, conical radar antenna scanning is considered in the case of circular polarization. Chapter 6 is devoted to the definition of fluctuations of the target return signal parameters caused simultaneously by the moving radar and radar antenna scanning. The correlation functions of the target return signal in space and surface scanning are defined and the problems associated with the moving radar with the line and conical radar antenna scanning are discussed. The problems of space and surface scanning with the Gaussian directional diagram are considered. The minimum radar range of navigational systems for the case of the Gaussian directional diagram is investigated. The sinc2-directional diagram is studied and the instantaneous and average power spectral densities of the target return signals are discussed. Theoretical study is strengthened by computer modeling and experimental results. Chapter 7 deals with the fluctuations of the target return signal parameters caused by moving reflectors of the radar antenna under the stimulus of the wind. The following cases are discussed: the deterministic motion of radar antenna reflectors under the stimulus of the layered wind, the stochastic motion and rotation of radar antenna reflectors, and the simultaneous deterministic and stochastic rotation of radar antenna reflectors.

Copyright 2005 by CRC Press

1598_C00.fm Page xiii Friday, October 8, 2004 2:00 PM

Chapter 8 is concerned with the study of the fluctuations of the target return signal parameters in navigational systems in scanning the two-dimensional (surface) target by the continuous frequency-modulated searching signal. Searching signals with linear frequency and the searching nonsymmetric saw-tooth frequency-modulated signals are discussed. The problems of scanning in definite angle and vertical scanning and moving are investigated. Additionally, the searching symmetric saw-tooth frequency-modulated signals and the searching harmonic frequency-modulated signals are studied. Phase characteristics of the target return signals in harmonic frequency modulation are discussed. Chapter 9 focuses on the study of fluctuations of the target return signal parameters in scanning the three-dimensional (space) target by the continuous searching signal with varying frequency. Nontransformed and transformed searching signals are considered. As a particular case, the nonperiodic and periodic frequency-modulated searching signals are investigated. The average power spectral density of the target return signal in the periodic frequency-modulated searching signal is defined. Chapter 10 discusses the problems of the target return signal parameter fluctuations caused by the change in the frequency from searching pulse to searching pulse. In the case of scanning the three-dimensional (space) target, the nonperiodic change in the frequency of the searching signals, the interperiodic fluctuations of parameters of the target return signals, the average power spectral density of the target return signal, and the periodic frequency modulation of searching signals are investigated. Moreover, the problems of scanning the two-dimensional (surface) target are discussed. The classification of stochastic target return signals is discussed. Chapter 11 focuses on the main theoretical principles of the generalized approach to signal processing in the presence of additive Gaussian noise. The basic concepts of the signal detection problem are discussed. The criticism of classical and modern signal processing theories from the viewpoint of defining the jointly sufficient mean and variance statistics of the likelihood function (or functional) is explored and modifications and initial premises of the generalized approach to signal processing in noise are considered. The likelihood function (or functional) possessing the jointly sufficient mean and variance statistics in the generalized approach to signal processing in noise is investigated. The engineering interpretation of the generalized approach to signal processing in noise is discussed and the model of the generalized detector in the cases of both slow and rapid fluctuating noise is studied. Chapter 12 is devoted to the main principles of the use of the generalized approach to the space–time signal and image processing. The basic concepts and foundations are considered. The problems of pattern recognition are discussed and the singularities of the generation of optical signals and radar images of the Earth’s surface are investigated. Chapter 13 focuses on the use of the generalized approach to the space–time signal and image processing in specific navigational systems.

Copyright 2005 by CRC Press

1598_C00.fm Page xiv Friday, October 8, 2004 2:00 PM

The generalized space–time signal and image processing algorithms are considered and compared with the classical correlation space–time signal and image processing algorithms. The difference generalized image processing algorithm is investigated and the generalized phase image processing algorithm is discussed. The invariant moments, amplitude ranking, gradient vector summing, structural methods, and bipartite functions are defined and investigated. The hierarchy generalized image processing algorithm is considered. The problems of the use of the more informative image area, coding of images, and superposition of point images are investigated in the use of the generalized image processing algorithm in specific navigational systems. The multichannel generalized image processing algorithm is discussed. Chapter 14 explores the use of the generalized approach in image preprocessing. The problems of image distortions, geometric transformations, image intensity distribution, detection of boundary edges, and sampling of images are discussed. The content of the book shows us that it is possible to raise higher the upper boundary of the potential noise immunity for complex and specific navigational systems in various areas of applications in the use of the generalized approach to signal and image processing in comparison with the noise immunity defined by classical and modern signal and image processing algorithms used in navigational systems.

Copyright 2005 by CRC Press

1598_C00.fm Page xv Friday, October 8, 2004 2:00 PM

Contents

Part I Theory of Fluctuating Target Return Signals in Navigational Systems Chapter 1 Probability Distribution Density of the Amplitude and Phase of the Target Return Signal Two-Dimensional Probability Distribution Density of the Amplitude and Phase 1.2 Probability Distribution Density of the Amplitude 1.3 Probability Distribution Density of the Phase 1.4 Probability Distribution Density Parameters of the Target Return Signal as a Function of the Distribution Law of the Amplitude and Phase of Elementary Signals 1.4.1 Uniform Probability Distribution Density of Phases 1.4.2 “Triangular” Probability Distribution Density of Phases 1.4.3 Gaussian Probability Distribution Density of Phases 1.5 Conclusions References 1.1

Chapter 2 Correlation Function of Target Return 2.1

2.2

2.3

Copyright 2005 by CRC Press

Signal Fluctuations Target Return Signal Fluctuations 2.1.1 Physical Sources of Fluctuations 2.1.2 The Target Return Signal: A Poisson Stochastic Process The Correlation Function and Power Spectral Density of the Target Return Signal 2.2.1 Space–Time Correlation Function 2.2.2 The Power Spectral Density of Nonstationary Target Return Signal Fluctuations The Correlation Function with the Searching Signal of Arbitrary Shape 2.3.1 General Statements 2.3.2 The Correlation Function with the Narrow-Band Searching Signal 2.3.3 The Correlation Function with the Pulsed Searching Signal 2.3.4 The Average Correlation Function

1598_C00.fm Page xvi Friday, October 8, 2004 2:00 PM

2.4

The Correlation Function under Scanning of the Three-Dimensional (Space) Target 2.4.1 General Statements 2.4.2 The Correlation Function with the Pulsed Searching Signal 2.4.3 The Target Return Signal Power with the Pulsed Searching Signal 2.4.4 The Correlation Function and Power of the Target Return Signal with the Simple Harmonic Searching Signal 2.5 The Correlation Function in Angle Scanning of the Two-Dimensional (Surface) Target 2.5.1 General Statements 2.5.2 The Correlation Function with the Pulsed Searching Signal 2.5.3 The Target Return Signal Power with the Pulsed Searching Signal 2.5.4 The Correlation Function and Power of the Target Return Signal with the Simple Harmonic Searching Signal 2.6 The Correlation Function under Vertical Scanning of the Two-Dimensional (Surface) Target 2.7 Conclusions References

Chapter 3 Fluctuations under Scanning of the 3.1

3.2

Copyright 2005 by CRC Press

Three-Dimensional (Space) Target with the Moving Radar Slow and Rapid Fluctuations 3.1.1 General Statements 3.1.2 The Fluctuations in the Radar Range 3.1.2.1 The Square Waveform Target Return Signal without Frequency Modulation 3.1.2.2 The Gaussian Target Return Signal without Frequency Modulation 3.1.2.3 The Smoothed Target Return Signal without Frequency Modulation 3.1.2.4 The Square Waveform Target Return Signal with Linear-Frequency Modulation 3.1.2.5 The Gaussian Target Return Signal with Linear-Frequency Modulation 3.1.3 The Doppler Fluctuations The Doppler Fluctuations of a High-Deflected Radar Antenna

1598_C00.fm Page xvii Friday, October 8, 2004 2:00 PM

3.2.1

The Power Spectral Density for an Arbitrary Directional Diagram 3.2.2 The Power Spectral Density for the Gaussian Directional Diagram 3.2.3 The Power Spectral Density for the Sinc-Directional Diagram 3.2.4 The Power Spectral Density for Other Forms of the Directional Diagram 3.3 The Doppler Fluctuations in the Arbitrarily Deflected Radar Antenna 3.3.1 General Statements 3.3.2 The Gaussian Directional Diagram 3.3.3 Determination of the Power Spectral Density 3.4 The Total Power Spectral Density with the Pulsed Searching Signal 3.4.1 General Statements 3.4.2 Interperiod Fluctuations in the Glancing Radar Range 3.4.3 Interperiod Fluctuations in the Fixed Radar Range 3.4.4 Irregularly Moving Radar 3.5 Conclusions References

Chapter 4 Fluctuations under Scanning of the Two4.1 4.2 4.3

4.4

Copyright 2005 by CRC Press

Dimensional (Surface) Target by the Moving Radar General Statements The Continuous Searching Nonmodulated Signal The Pulsed Searching Signal with Stationary Radar 4.3.1 General Statements 4.3.2 The Arbitrary Vertical-Coverage Directional Diagram: The Gaussian Pulsed Searching Signal 4.3.3 The Arbitrary Vertical-Coverage Directional Diagram: The Square Waveform Pulsed Searching Signal 4.3.4 The Gaussian Vertical-Coverage Directional Diagram: The Square Waveform Pulsed Searching Signal The Pulsed Searching Signal with the Moving Radar: The Aspect Angle Correlation Function 4.4.1 General Statements 4.4.2 The Arbitrary Vertical-Coverage Directional Diagram: The Gaussian Pulsed Searching Signal 4.4.3 The Gaussian Vertical-Coverage Directional Diagram: The Gaussian Pulsed Searching Signal 4.4.4 The Wide-Band Vertical-Coverage Directional Diagram: The Square Waveform Pulsed Searching Signal

1598_C00.fm Page xviii Friday, October 8, 2004 2:00 PM

4.5

The Pulsed Searching Signal with the Moving Radar: The Azimuth Correlation Function 4.5.1 General Statements 4.5.2 The High-Deflected Radar Antenna 4.5.3 The Low-Deflected Radar Antenna 4.6 The Pulsed Searching Signal with the Moving Radar: The Total Correlation Function and Power Spectral Density of the Target Return Signal Fluctuations 4.6.1 General Statements 4.6.2 The Gaussian Directional Diagram: The Gaussian Pulsed Searching Signal 4.6.3 The Gaussian Directional Diagram: The Square Waveform Pulsed Searching Signal 4.6.4 The Pulsed Searching Signal with Low Pulse Period-to-Pulse Duration Ratio 4.7 Short-Range Area of the Radar Antenna 4.8 Vertical Scanning of the Two-Dimensional (Surface) Target 4.8.1 The Intraperiod Fluctuations in Stationary Radar 4.8.2 The Interperiod Fluctuations with the Vertically Moving Radar 4.8.3 The Interperiod Fluctuations with the Horizontally Moving Radar 4.9 Determination of the Power Spectral Density 4.10 Conclusions References

Chapter 5 Fluctuations Caused by Radar 5.1

5.2

5.3

5.4

Copyright 2005 by CRC Press

Antenna Scanning General Statements 5.1.1 The Correlation Function under Space Scanning 5.1.2 The Correlation Function under Surface Scanning 5.1.3 The General Power Spectral Density Formula Line Scanning 5.2.1 One-Line Circular Scanning 5.2.2 Multiple-Line Circular Scanning 5.2.3 Line Segment Scanning 5.2.4 Line Circular Scanning with Various Directional Diagrams under Transmitting and Receiving Conditions Conical Scanning 5.3.1 Three-Dimensional (Space) Target Tracking 5.3.2 Two-Dimensional (Surface) Target Tracking Conical Scanning with Simultaneous Rotation of Polarization Plane

1598_C00.fm Page xix Friday, October 8, 2004 2:00 PM

5.5 Conclusions References

Chapter 6 Fluctuations Caused by the Moving Radar with Simultaneous Radar Antenna Scanning General Statements 6.1.1 The Correlation Function in the Scanning of the Three-Dimensional (Space) Target 6.1.2 The Correlation Function in the Scanning of the Two-Dimensional (Surface) Target 6.2 The Moving Radar with Simultaneous Radar Antenna Line Scanning 6.2.1 Scanning of the Three-Dimensional (Space) Target: The Gaussian Directional Diagram 6.2.2 Scanning of the Two-Dimensional (Surface) Target: The Gaussian Directional Diagram 6.2.3 Short-Range Area: The Gaussian Directional Diagram 6.2.4 The Sinc2-Directional Diagram 6.3 The Moving Radar with Simultaneous Radar Antenna Conical Scanning 6.3.1 The Instantaneous Power Spectral Density 6.3.2 The Averaged Power Spectral Density 6.4 Conclusions References 6.1

Chapter 7 Fluctuations Caused by Scatterers Moving 7.1

7.2

Copyright 2005 by CRC Press

under the Stimulus of the Wind Deterministic Displacements of Scatterers under the Stimulus of the Layered Wind 7.1.1 The Radar Antenna Is Deflected in the Horizontal Plane 7.1.2 The Radar Antenna Is Deflected in the Vertical Plane 7.1.3 The Radar Antenna Is Directed along the Line of the Moving Radar 7.1.4 The Stationary Radar Scatterers Moving Chaotically (Displacement and Rotation) 7.2.1 Amplitudes of Elementary Signals Are Independent of the Displacements of Scatterers 7.2.2 The Velocity of Moving Scatterers Is Random but Constant 7.2.3 The Amplitude of the Target Return Signal Is Functionally Related to Radial Displacements of Scatterers 7.2.4 Chaotic Rotation of Scatterers

1598_C00.fm Page xx Friday, October 8, 2004 2:00 PM

7.2.5

Simultaneous Chaotic Displacements and Rotations of Scatterers 7.3 Simultaneous Deterministic and Chaotic Motion of Scatterers 7.3.1 Deterministic and Chaotic Displacements of Scatterers 7.3.2 Chaotic Rotation of Scatterers and Rotation of the Polarization Plane 7.3.3 Chaotic Displacements of Scatterers and Rotation of the Polarization Plane 7.4 Conclusions References

Chapter 8 Fluctuations under Scanning of the Two-Dimensional (Surface) Target with the Continuous Frequency-Modulated Signal 8.1 General Statements 8.2 The Linear Frequency-Modulated Searching Signal 8.3 The Asymmetric Saw-Tooth Frequency-Modulated Searching Signal 8.3.1 Sloping Scanning 8.3.2 Vertical Scanning and Motion 8.3.3 Vertical Scanning: The Velocity Vector Is Outside the Directional Diagram 8.4 The Symmetric Saw-Tooth Frequency-Modulated Searching Signal 8.5 The Harmonic Frequency-Modulated Searching Signal 8.6 Phase Characteristics of the Transformed Target Return Signal under Harmonic Frequency Modulation 8.7 Conclusions References

Chapter 9 Fluctuations under Scanning of the 9.1 9.2

9.3

Copyright 2005 by CRC Press

Three-Dimensional (Space) Target by the Continuous Signal with a Frequency that Varies with Time General Statements The Nontransformed Target Return Signal 9.2.1 The Searching Signal with Varying Nonperiodic Frequency 9.2.2 The Periodic Frequency-Modulated Searching Signal 9.2.3 The Average Power Spectral Density with the Periodic Frequency-Modulated Searching Signal The Transformed Target Return Signal 9.3.1 Nonperiodic and Periodic Frequency-Modulated Searching Signals

1598_C00.fm Page xxi Friday, October 8, 2004 2:00 PM

9.3.2

The Average Power Spectral Density with the Periodic Frequency-Modulated Searching Signal 9.4 Conclusions References

Chapter 10 Fluctuations Caused by Variations in Frequency from Pulse to Pulse Three-Dimensional (Space) Target Scanning 10.1.1 Nonperiodic Variations in the Frequency of the Searching Signal 10.1.2 The Interperiod Fluctuations 10.1.3 The Average Power Spectral Density 10.1.4 Periodic Frequency Modulation 10.2 Two-Dimensional (Surface) Target Scanning 10.3 Conclusions References 10.1

Part II Generalized Approach to Space–Time Signal and Image Processing in Navigational Systems Chapter 11 Foundations of the Generalized Approach to Signal Processing in Noise Basic Concepts Criticism Initial Premises Likelihood Ratio The Engineering Interpretation Generalized Detector 11.6.1 The Case of the Slow Fluctuations 11.6.2 The Case of the Rapid Fluctuations 11.7 Conclusions References. 11.1 11.2 11.3 11.4 11.5 11.6

Chapter 12 Theory of Space–Time Signal and Image 12.1 12.2

Copyright 2005 by CRC Press

Processing in Navigational Systems Basic Concepts of Navigational System Functioning Basics of the Generalized Approach to Signal and Image Processing in Time 12.2.1 The Signal with Random Initial Phase 12.2.2 The Signal with Stochastic Amplitude and Random Initial Phase

1598_C00.fm Page xxii Friday, October 8, 2004 2:00 PM

12.3

Basics of the Generalized Approach to Space–Time Signal and Image Processing 12.4 Space–Time Signal Processing and Pattern Recognition Based on the Generalized Approach to Signal Processing 12.5 Peculiarities of Optical Signal Formation 12.6 Peculiarities of the Formation of the Earth’s Surface Radar Image 12.7 Foundations of Digital Image Processing 12.8 Conclusions References

Chapter 13 Implementation Methods of the Generalized Approach to Space–Time Signal and Image Processing in Navigational Systems 13.1 Synthesis of Quasioptimal Space–Time Signal and Image Processing Algorithms Based on the Generalized Approach to Signal Processing 13.1.1 Criterial Correlation Functions 13.1.2 Difference Criterial Functions 13.1.3 Spectral Criterial Functions 13.1.4 Bipartite Criterial Functions 13.1.5 Rank Criterial Functions 13.2 The Quasioptimal Generalized Image Processing Algorithm 13.3 The Classical Generalized Image Processing Algorithm 13.4 The Difference Generalized Image Processing Algorithm 13.5 The Generalized Phase Image Processing Algorithm 13.6 The Generalized Image Processing Algorithm: Invariant Moments 13.7 The Generalized Image Processing Algorithm: Amplitude Ranking 13.8 The Generalized Image Processing Algorithm: Gradient Vector Sums 13.9 The Generalized Image Processing Algorithm: Bipartite Functions 13.10 The Hierarchical Generalized Image Processing Algorithm 13.11 The Generalized Image Processing Algorithm: The Use of the Most Informative Area 13.12 The Generalized Image Processing Algorithm: Coding of Images 13.13 The Multichannel Generalized Image Processing Algorithm 13.14 Conclusions References

Copyright 2005 by CRC Press

1598_C00.fm Page xxiii Friday, October 8, 2004 2:00 PM

Chapter 14 Object Image Preprocessing 14.1 14.2

Object Image Distortions Geometrical Transformations 14.2.1 The Perspective Transformation 14.2.2 Polynomial Estimation 14.2.3 Transformations of Brightness Characteristics 14.3 Detection of Boundary Edges 14.4 Conclusions References

Appendix I Classification of Stochastic Processes References

Appendix II The Power Spectral Density of the Target Return Signal with Arbitrary Velocity Vector Direction of the Moving Radar in Space and with the Presence of Roll and Pitch Angles References Notation Index

Copyright 2005 by CRC Press

1598_Notation Index.fm Page 621 Friday, October 1, 2004 2:05 PM

Notation Index

A(t) a*(x, y, t) a(t) a(x, y, Λ, t) a*(t) alat(i∆x, j∆y) C(x), S(x) C(ωx , ωy) c da E(x, y)

(ω) –1(ω) f (S, φ) fs,ϑ(s, ϑ) f (S) f (x, y) f (φ) G0 G ( x , xz ) |g| g(ϕ, ψ) →

~

g(ϕ, ψ)

Copyright 2005 by CRC Press

signal amplitude factor model image signal information space–time signal model signal lattice function Fresnel integrals transfer function velocity of light diameter of the radar antenna lightness corresponding to the information signal in the moving image Fourier transform inverse Fourier transform two-dimensional probability distribution density of the amplitude and phase of the target return signal two-dimensional probability distribution density of the amplitude and phase of elementary signals probability distribution density of normalized amplitude of the target return signal two-dimensional probability distribution density probability distribution density of the phase of the target return signal amplifier coefficient of the radar antenna Green function gradient of velocity of the wind normalized two-dimensional directional diagram of the radar antenna generalized two-dimensional directional diagram of the radar antenna

1598_Notation Index.fm Page 622 Friday, October 1, 2004 2:05 PM

622 gh(ϕ) gt(ϕ, ψ) gr(ϕ, ψ) gv(ϕ) h h(x, y) H(x) (x, λx, y, λy) I, Ix, Iy Ik(x) J J0(x) ka kah kav kp kω L(Λ) λ (Λ) m0 n P PD PF PS Ptr Pfr

Copyright 2005 by CRC Press

Signal and Image Processing in Navigational Systems normalized radar antenna directional diagram by power in the horizontal plane normalized radar antenna directional diagram under the condition of transmission normalized radar antenna directional diagram under the condition of receiving normalized radar antenna directional diagram by power in the vertical plane altitude weight function Hermite polynomial function of scattering criterial functions modified Bessel function Jacobian Bessel function of the first order coefficient of the shape of the radar antenna directional diagram coefficient of the shape of horizontal-coverage directional diagram of the radar antenna coefficient of the shape of vertical-coverage directional diagram of the radar antenna coefficient of the pulsed searching signal shape velocity of frequency variation likelihood functional spectral energy brightness of the heated object likelihood function mean with respect to an ensemble of numbers of scatterers -dimensional noise vector probability probability of true detection probability of false alarm power of the searching signal probability of true location of the object image with reference to the control point probability of false location of the object image with reference to the control point

1598_Notation Index.fm Page 623 Friday, October 1, 2004 2:05 PM

Notation Index p Q(jω) q(ξ, ζ)

q2(ξ, ζ) R(t, τ) R(τ) R∆x(t1, t2) R∆ρ,∆x(t, τ, ∆ω)

R∆enρ, ∆x (t , τ , ∆ω )

R∆ρ, ∆ϕ , ∆ψ , ∆ξ , ∆ζ (t , τ)

R∆enρ, ∆ϕ , ∆ψ , ∆ξ , ∆ζ (t , τ)

R(∆x, ∆y, ∆t) Ri,j(x) R(t, τ) R(τ) Rg(∆, ∆β0, ∆γ0) Rp(τ, ∆) Rq(∆ξ)

Copyright 2005 by CRC Press

623 power of the target return signal frequency response of the linear system function representing the dependence of the amplitude of the target return signal on orientation of scatterer in space effective scattering area with fixed values of the angles ξ and ζ correlation function of fluctuations of the nonstationary target return signal correlation function of fluctuations of the stationary target return signal correlation function of space and time fluctuations of the nonstationary target return signal high-frequency correlation function of space and time fluctuations of the nonstationary target return signal envelope of high-frequency correlation function of space and time fluctuations of the nonstationary target return signal total correlation function of fluctuations of the target return signal caused by the moving radar, displacements and rotation of scatterers, and antenna scanning envelope of the total correlation function of the target return signal fluctuations caused by the moving radar, displacements and rotation of scatterers, and antenna scanning space–time correlation function criterial correlation function normalized correlation function of fluctuations of the nonstationary target return signal normalized correlation function of fluctuations of the stationary target return signal normalized correlation function of Doppler fluctuations of the target return signal normalized correlation function of rapid fluctuations of the target return signal normalized correlation function of space fluctuations of the target return signal caused by the rotation of the radar antenna polarization plane

1598_Notation Index.fm Page 624 Friday, October 1, 2004 2:05 PM

624 Rmov,sc(∆, ∆β0, ∆γ0)

Rmov(∆)

Rsc(∆β0, ∆γ0)

Rβ(∆, ∆β0) Rγ(∆, ∆γ0, t, τ) R∆x(t1, t2) R∆ρ,∆ϕ,∆ψ,∆ξ,∆ζ(t, τ)

R∆enρ ,∆ϕ ,∆ψ ,∆ξ ,∆ζ ( t , τ )

R∆ρ ,∆ϕ ,∆ψ ( t , τ )

R∆ξ ,∆ζ ( ∆ξ , ∆ζ )

S(t) S(ρ, x) S(ω) Sen(ω) S(ω, t) Sa*(ωx) SX(ωx)

Copyright 2005 by CRC Press

Signal and Image Processing in Navigational Systems total normalized correlation function of fluctuations of the target return signal caused by the moving radar with simultaneous radar antenna scanning normalized correlation function of fluctuations of the target return signal caused only by the moving radar normalized correlation function of space fluctuations of the target return signal caused only by radar antenna scanning azimuth normalized correlation function of fluctuations of the target return signal aspect-angle normalized correlation function of fluctuations of the target return signal correlation function of space and time fluctuations of the nonstationary target return signal total normalized correlation function of fluctuations of the target return signal caused by the moving radar, displacements and rotation of scatterers, and antenna scanning the envelope of the total normalized correlation function of fluctuations of the target return signal caused by the moving radar, displacements and rotation of scatterers, and antenna scanning particular normalized correlation function of fluctuations of the target return signal caused by the moving radar, displacement of scatterers, and antenna scanning particular normalized correlation function of fluctuations of the target return signal caused by the rotation of scatterers and polarization plane of the radar antenna amplitude of the searching signal amplitude of the received target return signal power spectral density of fluctuations of the target return signal envelope of the regulated power spectral density of fluctuations of the target return signal instantaneous power spectral density of fluctuations of the nonstationary target return signal power spectral density of the model image power spectral density of the moving image

1598_Notation Index.fm Page 625 Friday, October 1, 2004 2:05 PM

Notation Index

S (ω , t ) Sen(ω, t) Smov(ω, t)

Smov,sc(ω, t)

Ssc(ω, t)

S(ω, Ω) S(ω1, ω2) S(ωx, ωy) Sg(ω)

Sh(ω)

Sv(ω)

Sγ(ω)

Sβ(ω)

(ω) (ωx, ωy) si(t) Copyright 2005 by CRC Press

625 average power spectral density of the nonstationary target return signal envelope of the corrugated power spectral density of fluctuations of the target return signal continuous power spectral density of fluctuations of the target return signal caused only by the moving radar power spectral density of fluctuations of the target return signal caused by the moving radar with simultaneous radar antenna scanning regulated power spectral density of fluctuations of the target return signal caused only by radar antenna scanning two-dimensional power spectral density of fluctuations of the nonstationary target return signal two-dimensional power spectral density of fluctuations of the target return signal power spectral density of spatial frequency power spectral density of the Doppler fluctuations of the target return signal caused by variations in radial velocity within the limits of whole width of the radar antenna directional diagram power spectral density of the Doppler fluctuations of the target return signal caused by variations in radial velocity within the limits of the horizontal width of the radar antenna directional diagram power spectral density of the Doppler fluctuations of the target return signal caused by variations in radial velocity within the limits of the vertical width of the radar antenna directional diagram power spectral density of the Doppler fluctuations of the target return signal caused by variations in radial velocity within the limits of the aspect-angle plane of the radar antenna directional diagram power spectral density of the Doppler fluctuations of the target return signal caused by variations in radial velocity within the limits of the azimuth plane of the radar antenna directional diagram power spectral density of fluctuations of the target return signal shifted in frequency cross section of the power spectral density S(ωx, ωy) amplitude of the i-th elementary signal

1598_Notation Index.fm Page 626 Friday, October 1, 2004 2:05 PM

626 St S° S°N S(t) Ta Td Tp Tr Tsc t T(x) U(t, ω) Va Vr Vr

0

Vw Vw 0 Vwr W(t) W(x, t) Wh(t) Wtri(t) wi(t) w(x, t) X(t), Y(t) X X(x, y, t) α, θ α0 , θ0 β, γ

Copyright 2005 by CRC Press

Signal and Image Processing in Navigational Systems effective scattering area specific effective scattering area specific effective scattering area under vertical scanning normalized amplitude of the target return signal period of radar antenna hunting delay of the target return signal period of the pulsed signal effective duration of the target return signal period of radar antenna scanning observed instant of time Toronto function searching signal velocity of the moving radar relative to the Earth’s surface radial component of velocity of the moving radar projection of velocity of the moving radar on the axis of radar antenna directional diagram velocity of the wind velocity of the wind on the altitude h radial component of the velocity of scatterers target return signal resulting target return signal heterodyne signal transformed target return signal from the i-th scatterer i-th elementary signal elementary signal reflected by individual scatterer quadrature components of the target return signal m-dimensional vector of the observed signal at the input of the navigational system receiver moving image angles defining the position of scatterers in the polar coordinates angles defining the position of the axis of the radar antenna directional diagram in the polar coordinates azimuth and aspect angle of scatterer in the spherical coordinates

1598_Notation Index.fm Page 627 Friday, October 1, 2004 2:05 PM

Notation Index β0 , γ0 βes, γes ˜ β βab βsc βtub γ* ∆a ∆h ∆(h2) ∆v ∆(v2) ∆F, ∆Ω ∆Fd ∆Fmov

∆Fsc

∆ ∆β0, ∆γ0 ∆ζ, ∆ξ ∆ρ ∆ρ0 ∆ρϕ,ψ ∆ϕ, ∆ψ

Copyright 2005 by CRC Press

627 azimuth and aspect angle of the axis of the radar antenna directional diagram in the spherical coordinates azimuth and aspect angle of equisignal direction in the spherical coordinates coefficient of attenuation coefficient of molecular absorption coefficient of scattering by particles coefficient of scattering by nonhomogeneities caused by turbulence aspect angle of the center of observed radar range element effective width of radar antenna directional diagram effective width of horizontal-coverage directional diagram of the radar antenna effective width of square of horizontal-coverage directional diagram of the radar antenna effective width of vertical-coverage directional diagram of the radar antenna effective width of square of vertical-coverage directional diagram of the radar antenna effective bandwidth of the power spectral density of fluctuations of the target return signal effective bandwidth of the power spectral density of Doppler fluctuations of the target return signal effective bandwidth of the power spectral density of fluctuations of the target return signal caused only by the moving radar effective bandwidth of the power spectral density of fluctuations of the target return signal caused only by radar antenna scanning displacement of radar shifts of the axis of the radar antenna directional diagram angles of rotation of scatterers displacement of scatterers radial shift along the axis of the radar antenna directional diagram deviation of radial shifts for various scatterers within the limits of the radar antenna directional diagram angles of displacement of scatterers

1598_Notation Index.fm Page 628 Friday, October 1, 2004 2:05 PM

628 ∆ϕsc , ∆ψsc ∆ϕrm, ∆ψrm ∆ωM Λ ∆Λ δ(x) δx ε0 ελ η(x, y, t) ζ

θ θ0

λ λx , λy Λ Λ* µ, µ µ0 µb ξ

ξ(t) Π(t) ρ ρ∗

Copyright 2005 by CRC Press

Signal and Image Processing in Navigational Systems angles of displacement of scatterers caused by radar antenna scanning angles of displacement of scatterers caused by moving radar deviation in frequency error vector delta function elementary volume trajectory angle or angle between the velocity vector of the moving radar and the horizon coefficient of radiation additional reference noise vector angle defining the position of the scatterer in space relative to the polarization plane of the radar antenna and the direction of the beam angle between the velocity vector of the moving radar and the direction of the scatterer angle between the velocity vector of the moving radar and the axis of the radar antenna directional diagram wavelength views of the parameter vector Λ n-dimensional vector of parameters of the navigational object coordinates vector estimation of the vector of parameters of the navigational object coordinates scale factor coefficient of reflection coefficient of brightness angle defining the position of the scatterer in space relative to the polarization plane of the radar antenna and direction of the beam noise at the preliminary filter output of the generalized detector envelope of the high-frequency pulsed searching signal (video signal) radar range or distance between the radar antenna and scatterer distance between the center of pulse volume and radar

1598_Notation Index.fm Page 629 Friday, October 1, 2004 2:05 PM

Notation Index ρsr σ σ2 σ2(t) φ ϑ τp Φ(x) ϕ, ψ ϕ0 χ Ψ(t) Ω0 ΩM Ωmax Ωp Ωsc Ω(t) Ω(ϕ, ψ) Ωρ ω ω0 ωav ωh ωim <…> sinc (x)

Copyright 2005 by CRC Press

629 conditional boundary of the short range root mean square deviation variance variance of the stochastic process at the instant of time t phase of the target return signal phase of elementary signal duration of the pulsed searching signal error integral angles defining the position of scatterer relative to the axis of the radar antenna directional diagram random initial phase of the signal coefficient of asymmetry phase modulation law Doppler frequency corresponding to the center of pulse volume modulation frequency maximum Doppler frequency instantaneous frequency of the periodic pulsed searching signal angular velocity of radar antenna scanning instantaneous frequency Doppler frequency range finder frequency frequency of the target return signal carrier frequency of the signal averaged within the limits of the modulation period high frequency frequency of the heterodyne signal intermediate frequency mean sinc-function

1598_book.fm Page 1 Monday, October 11, 2004 1:57 PM

Part I

Theory of Fluctuating Target Return Signals in Navigational Systems

Copyright 2005 by CRC Press

1598_book.fm Page 29 Monday, October 11, 2004 1:57 PM

2 Correlation Function of Target Return Signal Fluctuations

2.1 2.1.1

Target Return Signal Fluctuations Physical Sources of Fluctuations

The target return signal, being a stochastic process, is subject to random variations in parameters, which are called fluctuations.1–3 As the target return signal is the sum of a large number of elementary signals (see Figure 2.1), sources of fluctuations can be considered as variations in the amplitude, phase, or frequency of elementary signals that lead to corresponding variations in these parameters in the resulting target return signal. For example, scatterers can move and rotate under the stimulus of the wind. Radial components of motion can give rise to phase changes in elementary signals. Tangential components of motion can give rise to amplitude changes in elementary signals if these changes are comparable with the width ∆a of the radar antenna directional diagram. Rotational components, if scatterers do not have spherical symmetry, can give rise to both amplitude and phase changes in elementary signals.4–6

∆a Vr′′

Vr′

θ′′ Vr′′ Vr′

θ′ V FIGURE 2.1 Doppler spectrum formation.

29 Copyright 2005 by CRC Press

1598_book.fm Page 30 Monday, October 11, 2004 1:57 PM

30

Signal and Image Processing in Navigational Systems

In the case of radar moving relative to the two-dimensional (surface) or three-dimensional (space) target too, phase changes in elementary signals occur. Let us suppose that the two-dimensional (surface) or three-dimensional (space) target has large angle dimensions, then various scatterers are observed at different angles within the directional diagram with respect to the direction of moving radar (see Figure 2.1). So, phase changes in elementary signals are the sources of fluctuations in the target return signal. Instead, we can also say that a moving radar can give rise to the Doppler shift in the frequency of elementary signals, because relative radial velocities of various scatterers differ within the searching area, due to differences in tracking angles (Figure 2.1).1 The received target return signal is not a simple signal and contains an entire frequency spectrum corresponding to the energy spectrum of radial components of various elementary scatterer velocities. Beats between various frequencies of the energy spectrum manifest themselves as target return signal fluctuations, which are called Doppler beats; the phenomenon is called the secondary Doppler effect.7 If the transmitter, receiver, or detector in navigational systems and scatterers are stationary, then fluctuations can arise due to radar antenna scanning or rotation of the radar antenna polarization plane, because both these give rise to amplitude changes in elementary signals (scanning and polarization fluctuations).8 Fluctuations of the received target return signal can be due to the nonstationary state of the searching frequency. Variations in frequency give rise to phase changes in elementary signals. Therefore, these variations are different for various scatterers and depend on the radar range. Unequal phase changes in elementary signals can give rise to fluctuations of the target return signal parameters (for example, frequency fluctuations). Peculiarities of the interaction between frequency fluctuations and target return signal Doppler fluctuations are discussed in more detail in References 9 to 11.

2.1.2

The Target Return Signal: A Poisson Stochastic Process

Target return signal fluctuations at the receiver or detector input in navigational systems can be very often considered as a Poisson stochastic process caused by superposition of nonstochastic (in shape) elementary signals arising at random instants of time.1,12 This is true, for example, when the twodimensional (surface) or three-dimensional (space) target is scanned by the pulsed searching radar signal. After each pulsed searching signal, the target return signal, containing a large number of elementary signals reflected by individual scatterers, comes in at the receiver or detector input in navigational systems. Thus, elementary signals are high-frequency pulses with the same shape and duration as the pulsed searching radar signal. Incoming pulses possess stochastic amplitudes and, what is more important, the receiver or detector input in navigational systems arrives at random times that depend on the position of scatterers in space or on the surface.

Copyright 2005 by CRC Press

1598_book.fm Page 31 Monday, October 11, 2004 1:57 PM

Correlation Function of Target Return Signal Fluctuations

31

Superposition of these deterministic (in shape) pulsed signals generates a Poisson stochastic process, which represents the target return signal fluctuations in the radar range. Thus, the response of the two-dimensional (surface) or three-dimensional (space) target to the pulsed searching radar signal is the interval of the stochastic process arising in the propagation of the target return signal.13,14 If the radar and scatterers are mutually stationary and the parameters of the radar equipment are stable, the target return signal caused by each pulsed searching signal is an exact copy of the previous target return signal, and the stochastic process becomes periodic (see the solid line in Figure 2.2a). In the case when the radar moves or radar antenna scans, the rigorous periodicity of the stochastic process is broken and the target return signal is shifted from period to period. The interperiod or slow fluctuations appear in contrast to the intraperiod or rapid fluctuations in the radar range (see the dotted line in Figure 2.2a).15,16 W

(a)

t t′

Tp

t ′ + Tp

2Tp

t ′ + 2Tp

3Tp

(b)

t 0

t′

t ′ + 2Tp t ′ + Tp

t ′ + 3Tp

FIGURE 2.2 (a) The intraperiod fluctuations; (b) the slow fluctuations.

Copyright 2005 by CRC Press

1598_book.fm Page 32 Monday, October 11, 2004 1:57 PM

32

Signal and Image Processing in Navigational Systems

Let us consider the instantaneous values of the target return signal at neighboring periods at the instants of time t, which are fixed with respect to the origin of the period. Then, the instantaneous values of the target return signal will have the shape of slowly fluctuating pulsed signals (see Figure 2.2b), the envelope of which is the stochastic process caused by the changing state of the system radar–scatterer. The totality of these instants of time { t′ , t′ + Tp ,..., t′ + nTp } ≡ t′

(2.1)

is called the time section according to Zukovsky et al.17 General interperiod and intraperiod fluctuations are shown in Figure 2.3. The interperiod fluctuations can be caused by other sources in addition to the moving radar and antenna scanning, for example, by the nonstationary state of frequency of the signal transmitter (or signal generator) in navigational systems, rotation of the radar antenna polarization plane, displacements of scatterers under the stimulus of the wind, and so on. W

nTp

4Tp

t ′ + 4Tp

3Tp

t ′ + 3Tp

2Tp

t ′ + 2Tp

Tp

t ′ + Tp t t′

FIGURE 2.3 The intraperiod and interperiod fluctuations.

Copyright 2005 by CRC Press

1598_book.fm Page 33 Monday, October 11, 2004 1:57 PM

Correlation Function of Target Return Signal Fluctuations

33

(a)

(b) FIGURE 2.4 Elementary signals: (a) amplitude modulation; (b) amplitude–frequency modulation.

Target return signal fluctuations caused by the moving radar and antenna scanning with the searching simple signal can be called slow fluctuations. Unlike interperiod fluctuations, slow fluctuations also generate a Poisson stochastic process. The sources of fluctuations are the elementary signals reflected from elementary scatterers, which are modulated by the radar antenna directional diagram. Under scanning, this modulation is a pure amplitude modulation (see Figure 2.4a), whereas with moving radar, this modulation is an amplitude–frequency modulation (see Figure 2.4b). Therefore, frequency changes are caused by the Doppler shift, which is proportional to the radial component of scatterer relative velocity and varies during radar motion due to changes in the scanning angle. The target return signal is the sum of a large number of equivalent elementary signals that are nonstochastic (in shape) but arise at random instants of time.18-20 The Poisson stochastic process as a function of time can be determined by the sum of the deterministic functions w(t, ti) = siw(t – ti) with random parameters si and ti:

()

W t =

∞

∑

i = −∞

( )

w t , ti =

∞

∑ s w (t − t ) , i

i

(2.2)

i = −∞

where si is the amplitude of the i-th elementary signal; ti is the instant of time when the i-th elementary signal has arisen; t is the observed instant of time. The parameters si and ti are statistically independent random variables. Let us assume that the time instants of individual elementary signal appearances are independent events and that the probability of appearance of a Copyright 2005 by CRC Press

1598_book.fm Page 34 Monday, October 11, 2004 1:57 PM

34

Signal and Image Processing in Navigational Systems

single elementary signal within the limits of an infinitesimal time interval is proportional to the length of the time interval. In other words, the probability of appearance of n elementary signals within the limited time interval ∆t obeys the Poisson probability distribution law.12,21,22 Representation of the target return signal in navigational systems, in the form of superposition of elementary signals initiated with irregularity, is appropriate for two reasons. First, this representation allows us to define the statistics of the incoming target return signal at the receiver or detector input in navigational systems. Second, this representation also allows us to define the correlation function and spectral power density of the input target return signal fluctuations. If the average number of elementary signals arising within the time interval equal to the duration of the elementary signal is sufficiently high, then the target return signal can be considered as a Gaussian process, in a narrow sense. In other words, the probability distribution densities of any order are Gaussian. The shape of elementary signals does not play any role; in particular, elementary signals can be considered as pulses with a high radio-frequency carrier. This condition implies that a sufficiently large number of elementary targets must be within the scanning area, and the region filled by scatterers must be larger than the scanning area, so with scanning area displacement, some scatterers enter this area and others leave it, forming in this way a process of superposition of elementary signals.23–25 It is significant that in this case, displacements of individual scatterers are not assumed to be independent. Only the time instants of initiation of elementary signals can be considered to be independent. Displacements of scatterers can be correlated or may even be hardly dependent on each other, for example, in the case of moving radar or antenna scanning. Although the target return signal is considered Gaussian in a narrow sense, complete information about it is held in the correlation function or in the corresponding power spectral density of the target return signal fluctuations. The first part of this book is devoted to the investigation of the correlation function and power spectral density of target return signal fluctuations at the receiver or detector input in navigational systems. From Equation (2.2), we are able to define the main features of the Poisson stochastic process characterizing the target return signal. The main relationships are given here without proofs and will be used at a later time. If elementary signals are written in the complex-valued form, the correlation function of target return signal fluctuations can be determined as follows:12,21

()

R τ = n1 s

∞

2

∫ w (t ) w (t + τ) dt = R ( 0) ⋅ R (τ ) *

(2.3)

−∞

where n1 is the number of elementary signals per time; * denotes a complex conjugate value;

Copyright 2005 by CRC Press

1598_book.fm Page 35 Monday, October 11, 2004 1:57 PM

Correlation Function of Target Return Signal Fluctuations ∞

()

p = R 0 = n1 s

2

∫ w (t ) dt 2

35

(2.4)

−∞

is the power of the stochastic process; ∞

∫ w (t ) w (t + τ) dt *

()

R τ =

−∞

∞

∫ w (t )

(2.5) 2

dt

−∞

is the normalized correlation function. Using the Fourier transform in Equation (2.3) and the convolution theorem, we can define the power spectral density:23,26

( )

S ω =

p π

2

∞

∫ w (t ) ⋅ e

− jωt

−∞

∞

∫ w (t )

dt .

2

(2.6)

dt

−∞

The power spectral density obtained in Equation (2.6) coincides in shape with any individual elementary signal, clearly because all elementary signals are the same and differ only in amplitude and the time instant of initiation. Furthermore, their power spectral densities are identical and only the coefficient of proportionality and phase factor differ. The resulting power spectral density, equal to the average sum of a large number of identical elementary signals, coincides with the power spectral density of an individual elementary signal.27 From the previous considerations, it follows that with the target return signal represented as a Poisson stochastic process, the high-frequency pulsed signal (radio pulse) and the track of the directional diagram moving along the two-dimensional (surface) or three-dimensional (space) target with the velocity of radar antenna scanning or with the velocity of moving radar can play the role of an elementary signal. In the first case, if time is an independent variable of the stochastic process (time fluctuations), then in the second case, the angle and linear coordinates characterizing the position of the scanning area in the scattering environment are also independent variables of the stochastic process (space fluctuations). As all these cases can take place simultaneously, a multidimensional definition of elementary signals is necessary (instead of the one-dimensional definition) to generalize the concept of the pulsed stochastic process for the multidimensional space.1,27 Copyright 2005 by CRC Press

1598_book.fm Page 36 Monday, October 11, 2004 1:57 PM

36

2.2 2.2.1

Signal and Image Processing in Navigational Systems

The Correlation Function and Power Spectral Density of the Target Return Signal Space–Time Correlation Function

Let an elementary signal reflected by an individual scatterer at the fixed instant of time t be in the following form: w(x, t) = w(x1, x2, …, xn, t),

(2.7)

where x {x1, …, xn} is the set of variables describing the mutual position of the transmitter, receiver, or detector in navigational systems, and the scatterer. The function given by Equation (2.7) takes into consideration all essential factors: the shape and orientation of the transmitted and received radar antenna directional diagram, the position of the scatterer, the shape of the searching signal, and so on. The amplitude and phase of an elementary signal depend on these factors.28,29 Let us consider the n-dimensional space, where the variables x form any coordinate system. This space is then divided by coordinate surfaces into a large number of n-dimensional elementary volumes, the dimensions of which are infinitesimal, so that we can neglect the variation of the function w(x, t) within each volume; but there are a great number of scatterers in each volume. The target return signal of the i-th elementary volume can be written in the form: wi = miw(xi, t), where mi is the number of scatterers inside the elementary volume and xi = {x1i, x2i, …, xni} are the relative coordinates of the elementary volume. Let us assume that the resulting target return signal determined by

( )

W x, t =

∞

∑ m w (x , t) i

i

(2.8)

i=0

is a uniform field, which can be nonstationary.30 At the instant of time t = t1, the target return signal can be written in the following form:

( )

W1 x , t =

∞

∑ m w (x , t ) . i

i

1

(2.9)

i=0

If the variables xi change in value by some differential ∆x as a consequence of the moving directional diagram, then the target return signal can be written in the following form:

Copyright 2005 by CRC Press

1598_book.fm Page 37 Monday, October 11, 2004 1:57 PM

Correlation Function of Target Return Signal Fluctuations

( )

W2 x , t =

37

∞

∑ m w ( x + ∆x, t ) . i

i

(2.10)

2

i=0

In the general case, the differentials ∆x differ for various scatterers and are functions of the coordinate xi: ∆x {∆x1(xi), ∆x2(xi), …, ∆xn(xi)}.

(2.11)

The correlation function of the space and time fluctuations has the following form:

][

[

]

R∆x (t1 , t2 ) = W1 (x , t) − W1 (x , t) W2* (x , t) − W2* (x , t) ,

(2.12)

where ∞

∑ m w (x , t ) ;

(2.13)

∑ m w ( x + ∆x, t ) ;

(2.14)

( )

W1 x , t =

( )

W2 x , t =

i

i

1

i=0

∞

i

i

2

i=0

and the index ∆x indicates that the correlation function given by Equation (2.12) is determined for the fluctuations that are functionally related to changes in the variable x. Substituting Equation (2.13) and Equation (2.14) in Equation (2.12), we can write  ∞ R∆x t1 , t2 =  mi − mi w x i , t1  i=0 

(

∑(

)

) (

∞

=



∞

) ∑ ( m − m ) w * ( x j

  j=0

j

j

 + ∆x , t2   

)

,

∑ ( m − m ) w ( x , t ) w * ( x + ∆x, t ) 2

i

i

i

1

i

2

i=0

(2.15) because (mi – mi ) · w(xi, t) and (mj – m j ) · w(xj, t) are independent random variables under the condition i ≠ j. The total sum is equal to zero under the condition i ≠ j. Here the value (mi − mi ) is the variance of the number of scatterers at the i-th elementary volume. 2

Copyright 2005 by CRC Press

1598_book.fm Page 38 Monday, October 11, 2004 1:57 PM

38

Signal and Image Processing in Navigational Systems

Let us assume that the number of scatterers in any fixed elementary volume obeys the Poisson distribution law, i.e., the probability that any scatterer can appear in a given elementary volume depends on the dimensions of the volume and is independent of the position of the elementary volume in space. Then we can write ( m i − m i ) 2 = m i = m 0 × δx, where m0 = const– is the mean (with respect to an ensemble) of the number of scatterers per unit volume; and δx is the elementary volume.31 Going to the limit as δx → 0 and replacing the sum with the integral in Equation (2.15), we can write

(

)

R∆x t1 , t2 = m0

∞

∫ w ( x, t ) ⋅ w * ( x + ∆x, t ) dX , 1

2

(2.16)

−∞

where dX = |J| dx1 … dxn; J is the Jacobian corresponding to the chosen coordinate system, and integration is carried out over the whole n-dimensional domain, where the integrand differs from zero. In the general case, if the stochastic process is nonstationary, the result of the integration depends on the variables t1 and t2. If the stochastic process is stationary, the result of the integration is defined by the difference τ = t2 – t1. In the general case, the differentials ∆x, as was noted, are not the same for all scatterers. The differentials ∆x can be fixed-point or stochastic functions of multidimensional space coordinates. When the differentials ∆x are random, as in fluctuations caused by the stimulus of the wind, Equation (2.16) must be additionally averaged in accordance with the multidimensional probability distribution density of the differentials ∆x: R∆x (t1 , t2 ) =

∫R

∆x

(t1 , t2 ) f (∆x) d(∆x) ,

(2.17)

where f(∆x) = f(∆x1, …, ∆xn) is the multidimensional probability distribution density of the differentials ∆x and d(∆x) = d(∆x1) … d(∆xn) Equation (2.16) and Equation (2.17) are space–time correlation functions, because they represent correlation characteristics of the received target return signals with respect to both the time and space coordinates defining the geometry of the radar–scatterer system. The correlation function given by Equation (2.16) is functionally related with the space–time (frequency) power spectral density of the field using the multidimensional Fourier transform with respect to the coordinates x, t1, and t2.30 Using space coordinates, this Fourier transform gives us the ndimensional space power spectral density. Using time coordinates, we obtain the generalized two-dimensional power spectral density that takes into consideration the correlation relationships between the spectral power densities at frequencies ω1 and ω2. Henceforth, the space variables ∆x defining the dynamics of the radar–scatterer system will be represented by functions of

Copyright 2005 by CRC Press

1598_book.fm Page 39 Monday, October 11, 2004 1:57 PM

Correlation Function of Target Return Signal Fluctuations

39

time. Because of this, the space–time fluctuations can be represented by time fluctuations only, to define which we use the time correlation function and the frequency–time power spectral density. The nonstationary correlation function given by Equation (2.16) can be written in the following form: R∆x(t1, t2) = σ(t1) · σ(t2) · R∆x(t1, t2),

(2.18)

where σ 2 ( t i ) = m0

∫

2

w(x , ti ) dX ,

i = 1, 2

(2.19)

is the variance (power) of the stochastic process at the instants of time t1 and t2;

(

)

R∆x t1 , t2 =

( ) = σ (t ) ⋅ σ (t ) R∆x t1 , t2 1

2

∫ w ( x, t ) ⋅ w * ( x + ∆x, t ) dX ∫ w ( x, t ) dX ⋅ ∫ w ( x, t ) dX 1

2

2

1

2

(2.20)

2

is the nonstationary normalized correlation function. Equation (2.18)–Equation (2.20) are extensions of Equation (2.3)–Equation (2.5). In the analysis of nonstationary stochastic processes, we use the instantaneous correlation function in parallel with the correlation function given by Equation (2.16):

∫

R∆x (t , τ) = m0 w(x , t − 0.5τ) ⋅ w * (x + ∆x , t + 0.5τ)dX

(2.21)

results from Equation (2.16) by using the following transformations: t1 = t – 0.5τ;

(2.22)

t2 = t + 0.5τ;

(2.23)

τ = t2 – t1;

(2.24)

t = 0.5(t1 + t2).

(2.25)

Equation (2.16) can be written in the more symmetric form:

∫

R∆x (t , τ) = m0 w(x ′ − 0.5∆x , t – 0.5τ) ⋅ w * (x ′ + 0.5∆x , t + 0.5τ)dX , (2.26)

Copyright 2005 by CRC Press

1598_book.fm Page 40 Monday, October 11, 2004 1:57 PM

40

Signal and Image Processing in Navigational Systems

where x′ = x + 0.5∆x. Due to uniformity of the considered stochastic field, Equation (2.21) and Equation (2.26) are equivalent; so, the symbol “′” of the variable x in Equation (2.26) can be omitted. The functions given by Equation (2.16), Equation (2.21), and Equation (2.26) are equivalent too, to fit the stationary stochastic process. The transformation of Equation (2.16) in Equation (2.21) or Equation (2.26), i.e., the transformation of coordinates determined by Equation (2.22)–Equation (2.25) from the plane (t1, t2) into the plane (t, τ) rotated at 45°, allows us to separate, wherever possible, the stationary and nonstationary components of the stochastic process. When the target return signal is stationary, the correlation function R(t, τ) is independent of time t and becomes the ordinary stationary correlation function R(τ). In spite of the fact that the correlation function determined by Equation (2.21) contains complete information regarding both the power (or variance) of the target return signal and characteristics of the power spectral density as a function of time, it is convenient to separate the correlation function into components having the following form: R ∆x ( t , τ ) = R ∆ x ( t , 0 ) ∆ x = 0 ⋅ R ∆ x ( t , τ ) ,

(2.27)

where R∆x (t , 0) ∆x =0 = p(t) = σ 2 (t) = m0

∫ w(x, t) dX 2

(2.28)

and

R ∆x ( t , τ ) =

R ∆x ( t , τ ) = σ 2 (t )

∫ w(x, t − 0.5τ) ⋅ w * (x + ∆x, t + 0.5τ) dX . ∫ w(x, t) dX 2

(2.29)

The normalized correlation function given by Equation (2.29), as well as the correlation function in Equation (2.26) can be written in a more symmetric form. The peculiarity of the nonstationary target return signal when the variables t and τ of the correlation function are separated, i.e., when the correlation function is separated according to12 R(t, τ) = R1(τ) · R2(t),

(2.30)

is of prime interest to us. The power (or variance) of the target return signal depends on time: R(t, 0) = R1(0) × R2(t). The normalized correlation function and characteristics of the power spectral density are independent of time:

Copyright 2005 by CRC Press

1598_book.fm Page 41 Monday, October 11, 2004 1:57 PM

Correlation Function of Target Return Signal Fluctuations

R(t , τ) =

41

R1 ( τ) . R1 (0)

(2.31)

In cases where there is no need to get information regarding the power (or variance) of the target return signal, we can limit the study only to the normalized correlation function.32,33 Henceforth, in the majority of cases, the normalized correlation function will be written in a simpler form. For example, instead of Equation (2.29) we can write

( )

∫ (

)

(

)

R∆x t , τ = N w x , t − 0.5τ ⋅ w * x + ∆x , t + 0.5τ dX ,

(2.32)

where N=

1

∫ ( ) w x, t

2

(2.33) dX

is unity divided by the normalized correlation function of the target return signal fluctuations under the condition ∆x = τ = 0. Specific formulae for N are different and depend on the form of the normalized correlation function.

2.2.2

The Power Spectral Density of Nonstationary Target Return Signal Fluctuations

Because in the general case, the nonstationary correlation function of target return signal fluctuations is a function of two variables t1 and t2 or t and τ, the power spectral density of target return signal fluctuations, defined by the Fourier transform with respect to the variable τ, is a function depending both on frequency and on time. This statement does not agree with the usual concept of the power spectral density as a sum of harmonic elementary signals independent of time. Because of this, the more general definition of the power spectral density, which clearly allows us to determine it at the output of a linear system, uses the following form Sout(ω) = |Q(jω)|2 · Sin(ω)

(2.34)

or an analogous form, where Q(jω) is the frequency response of the linear system, and is introduced to fit nonstationary stochastic processes.34-36 The concept of the two-dimensional (generalized) power spectral density S(ω1, ω2),6,8 which is functionally related to the nonstationary correlation function R(t1, t2) by the following Fourier transforms

Copyright 2005 by CRC Press

1598_book.fm Page 42 Monday, October 11, 2004 1:57 PM

42

Signal and Image Processing in Navigational Systems

(

)

∞ ∞

S ω1 , ω 2 =

∫ ∫ R (t , t ) ⋅ e 1

2

(

j ω1t1 − ω 2t2

) dt dt , 1 2

(2.35)

−∞ −∞

where

(

)

R t1 , t2 =

1 4π 2

∞ ∞

∫ ∫ S (ω , ω ) ⋅ e 1

2

(

− j ω1t1 − ω 2t2

) dω dω , 1 2

(2.36)

−∞ −∞

is universally adopted. We can show that the generalized power spectral density is the mean of the product of the power spectral densities that are shifted in frequency, S(ω1) and S(ω2): S(ω 1 , ω 2 ) = S(ω 1 ) ⋅ S(ω 2 ) ,

(2.37)

where S(ω) is the Fourier transform of a sample of the stochastic process x(t), the nonstationary correlation function of which has the following form: R(t1 , t2 ) = x(t1 ) ⋅ x(t2 ) .

(2.38)

Comparing Equation (2.37) and Equation (2.38), one can see that the nonstationary power spectral density of the stochastic process x(t) is the nonstationary correlation function of the stochastic process S(ω) in frequency space. In the general case, this correlation function differs from zero. This circumstance indicates that there is a correlation function between the power spectral densities at the frequencies ω1 and ω2. As is well known, any frequencies satisfying the condition ω1 ≠ ω2 are not correlated to the power spectral density of the stationary stochastic process. The statement that a stationary stochastic process becomes nonstationary when there is an interspectral correlation relationship, is clear on the basis of physical reasoning. For example, if the stationary stochastic process is modulated in amplitude by the frequency Ω, then this process becomes nonstationary and all pairs of the spectral components not tuned on the frequency Ω exchange all their side-lobe components containing information about the amplitude and phase. Due to this fact, there is a correlation between these side-lobe spectral components. This correlation is strong, and the coefficient of modulation is high. Actually, the analogous effect can occur under frequency and phase modulation. If the modulation law is more complex, in particular, nonperiodic, then there is an exchange of spectral side-lobe components between all components of the initial power spectral density and between all correlated frequencies that the power spectral density can give rise to. So, the power spectral density will be fuzzy on the plane Copyright 2005 by CRC Press

1598_book.fm Page 43 Monday, October 11, 2004 1:57 PM

Correlation Function of Target Return Signal Fluctuations

43

(ω1, ω2). In the case of the stationary stochastic process, we can write S(ω1, ω2) = S(ω1) × δ(ω1 – ω2). In this case, the domain of the power spectral density definition contracts to the line ω1 = ω2. Furthermore, it is obvious that the power spectral densities are not correlated under the condition ω1 ≠ ω2. The concept of the instantaneous (time-frequency) power spectral density of the nonstationary stochastic process,12 which is the Fourier transform of the instantaneous correlation function R(t, τ) with respect to the variable τ ∞

( )

S t, ω =

∫ R (t, τ ) ⋅ e

− jωτ

dτ ,

(2.39)

−∞

is the second universally accepted concept, which is very convenient for theoretical investigations. The inverse Fourier transform in the form ∞

( )

1 R t, τ = 2π

∫ S (t, ω ) ⋅ e

jωτ

dω

(2.40)

−∞

with the condition τ = 0 allows us to write

( )

()

R t, 0 = x2 t =

1 2π

∞

∫ S (t, ω ) dω .

(2.41)

−∞

In other words, the instantaneous power spectral density defines the probability distribution law of power (or variance) of the stochastic process x(t) in the coordinate system (t, ω). The integral over all frequencies gives the mean x 2 (t) . With some values of ω and t, the power spectral density S(t, ω) can be negative.12 The instantaneous power spectral density S(t, ω), depending on the parameter t, yields very important information about the character of the stochastic process as a function of time. It is precisely this power spectral density, as a rule, that will subsequently be determined in the study of nonstationary stochastic processes. The deterministic dependence of the instantaneous power spectral density as a function of time can be shown in the frequency region if we are able to define the Fourier transform for the power spectral density S(t, ω) with respect to the parameter t:

(

)

S ω, Ω =

∞

∫ S (t, ω ) ⋅ e

−∞

Copyright 2005 by CRC Press

− jΩt

dt .

(2.42)

1598_book.fm Page 44 Monday, October 11, 2004 1:57 PM

44

Signal and Image Processing in Navigational Systems

One can see that the formula in Equation (2.42) is the double Fourier transform of the correlation function R(t, τ) of fluctuations of the target return signal

(

)

∞ ∞

S ω, Ω =

∫ ∫ R (t, τ ) ⋅ e

(

− j ωτ + Ωt

) dτ dt .

(2.43)

−∞ −∞

From previous considerations, the two-dimensional power spectral density determined by Equation (2.43) is equivalent to the power spectral density given by Equation (2.35) with coordinates ω and Ω related functionally to the coordinates ω1 and ω2 by relationships that are analogous to the ones given in Equation (2.22)–Equation (2.25): Ω = ω1 – ω2;

(2.44)

ω = 0.5(ω1 + ω2);

(2.45)

ω1 = ω – 0.5Ω;

(2.46)

ω2 = ω + 0.5Ω.

(2.47)

The relationships determined by Equation (2.44)–Equation (2.47) are a rotation of the coordinate axes at an angle equal to 45°. Substituting Equation (2.44)–Equation (2.47) in Equation (2.43), we can obtain Equation (2.35). In an analogous way [see Equation (2.22)–Equation (2.25)], the correlation functions R(t, τ) and R(t1, t2) [see Equation (2.36)] are functionally related. In the case of the stationary target return signal, the two-dimensional power spectral density S(ω, Ω) is equal to zero within the plane (ω, Ω), except for the line Ω = 0. This is the one-dimensional power spectral density. The appearance of a nonstationary state leads to the spreading of the power spectral density with respect to the coordinate Ω, giving rise to the twodimensional power spectral density. If the correlation function is separable [see Equation (2.30)], then we can write

(

)

S ω, Ω =

∞ ∞

∫ ∫ R ( τ ) R (t ) ⋅ e 1

(

− j ωτ + Ωt

2

( ) ( )

) dτ dt = S ω ⋅ S Ω , 1 2

(2.48)

−∞ −∞

where

( )

S1 ω =

∞

∫ R (τ) ⋅ e 1

−∞

Copyright 2005 by CRC Press

− jωτ

dt

(2.49)

1598_book.fm Page 45 Monday, October 11, 2004 1:57 PM

Correlation Function of Target Return Signal Fluctuations

45

and

( )

S2 Ω =

∞

∫ R (t ) ⋅ e

− jΩτ

2

dt .

(2.50)

−∞

Obviously, the instantaneous power spectral density S(ω, t) given by Equation (2.39) can be written in the form: S(t, ω) = S1(ω) · R2(t). In the case of the stationary target return signal under the condition R2(t) = const, we can write S2(Ω) = δ(Ω), and if R(t, τ) = const, we obtain S(ω, Ω) = S1(ω) · δ(Ω). This fact proves that in the case of the stationary target return signal, the power spectral density S(ω, Ω) is defined only along the axis Ω = 0 and has the shape of the power spectral density S1(ω). Peculiarities of the correlation functions and power spectral densities of nonstationary target return signal fluctuations are discussed in more detail in References 2, 21, and 37.

2.3

The Correlation Function with the Searching Signal of Arbitrary Shape

2.3.1

General Statements

Let us consider for simplicity that a single-position aircraft radar navigational system is generating the searching signal U(t, ω).38,39 Then, the target return signal from an individual scatterer takes the following form:

(

w(x , t) = S(ρ, x 2 , … , xn ) ⋅ U t −

2ρ c

)

(

, ω = S(ρ, x) ⋅ U t −

2ρ c

)

,ω ,

(2.51)

where x {x2, …, xn};

(2.52)

S(ρ, x) is the amplitude of the received target return signal; ρ is the radar range (in the case of navigational systems, ρ is the distance between the radar antenna and scatterer); and c is the velocity of light. The amplitude of the received target return signal depends on the distance ρ, mutual positions of the radar, scatterer, shape and orientation of the radar antenna directional diagram, position of the radar antenna polarization plane, effective scattering area of the scatterer, position of the scatterer in space, etc. This dependence is represented a function of the arguments ρ and x. Here and subsequently, the coordinate of the distance x1 ρ is extracted from the general totality, because the coordinate x1 has the special property of being a component of two factors simultaneously [see Equation (2.52)]. Copyright 2005 by CRC Press

1598_book.fm Page 46 Monday, October 11, 2004 1:57 PM

46

Signal and Image Processing in Navigational Systems

With the fixed parameter t, the target return signal at the radar receiver or detector input in navigational systems is the sum of elementary signals received from all scatterers at random points of the n-dimensional space within the scanned area. This target return signal depends on the time

(

parameter t, because the function U t −

2ρ c

)

, ω [see Equation (2.51)] slides

along the distance axis x1 ρ with a velocity equal to half the velocity of light as a result of changing the parameter t. Because the amplitude S(ρ, x2, …, xn) of the target return signal depends also on the radar range ρ (or the distance between the radar antenna and scatterer), the functional dependence on the time parameter t leads to the nonstationary state of the stochastic process, in the general case. Substituting Equation (2.51) in Equation (2.26), we can obtain the general formula for the space–time correlation function in the symmetrical form:

(

)

∫ ∫ S (ρ − 0.5∆ρ, x − 0.5∆x) ⋅ S (ρ + 0.5∆ρ, x + 0.5∆x) × U ( t − − 0.5τ + , ω − 0.5 ∆ω ) , × U * ( t − + 0.5τ − , ω + 0.5 ∆ω ) dρ dX

R∆ρ, ∆x t , τ , ∆ω = m0

2ρ c

∆ρ c

2ρ c

∆ρ c

(2.53) where ∆x = {∆x2, …, ∆xn} and dX = |J|dx2…dxn. 2.3.2

The Correlation Function with the Narrow-Band Searching Signal

As a rule, the searching signals generated by radar in navigational systems have a moderately narrow power spectral density with respect to the carrier frequency. So, these searching signals can be written in the following form:40,41 t

U (t) = S(t) ⋅ e

− j ω ( t ) dt

∫ 0

− j ω t+Ψ t = S(t) ⋅ e [ 0 ( )] ,

(2.54)

where ω(t) = ω0 + Ω(t); Ω(t) = dt( ) is the instantaneous frequency; and Ψ(t) is the phase modulation law, which is not taken into consideration by the term ω0t. Substituting Equation (2.54) in Equation (2.53), we can write dΨ t

R∆ρ,∆x (t , τ , ∆ω ) = R∆enρ ,∆x (t , τ , ∆ω ) ⋅ e jω 0τ , where

Copyright 2005 by CRC Press

(2.55)

1598_book.fm Page 47 Monday, October 11, 2004 1:57 PM

Correlation Function of Target Return Signal Fluctuations

(

)

47

∫∫ S (ρ − 0.5∆ρ, x − 0.5∆x) ⋅ S (ρ + 0.5∆ρ, x + 0.5∆x) × S ( t − − 0.5τ + ) ⋅ S ( t − + 0.5τ − ) ⋅ .

R∆enρ, ∆x t , τ , ∆ω = m0

2ρ c

×e

∆ρ c

∆ρ c

2ρ c

2ρ ∆ρ 2ρ ∆ρ   j  Ψ  t − + 0.5 τ −  − Ψ  t − 0.5 τ +    c c  c c   

⋅e

∆ρ 2ρ   − j 2ω ⋅ ∆ω  t −    c c  

dp dX (2.56)

The formula in Equation (2.56) defines the envelope of the high-frequency correlation function of space and time fluctuations of the nonstationary target return signal. The correlation function R∆enρ,∆x (t, τ, ∆ω) in Equation (2.56), with respect to the delay (radar range or distance) and the Doppler frequency (radial velocity),42 can be considered as an extension of the correlation function used in the theory of communications to define a signal passing through a channel with two-dimensional scattering, and the amplitude S(ρ,x) is the multidimensional function of scattering with respect to the coordinates ρ and x. In this case, the amplitude S(ρ,x) is reduced to the two-dimensional function of scattering with respect to the delay and Doppler frequency. The high-frequency correlation function R∆ρ,∆x(t, τ, ∆ω) given by Equation (2.55), as well as by Equation (2.54), consists of two factors: the rapidly varying function ejω0τ and the slowly varying function R∆enρ,∆x (t, τ, ∆ω) [see Equation (2.56)] that is the complex envelope of the correlation function or the low-frequency correlation function of the target return signal fluctuations. It is precisely this envelope R∆enρ, ∆x (t, τ, ∆ω) of the high-frequency correlation function that is of prime interest to us, because it characterizes features of the fluctuations. The cofactor ∆ρ

e

−2 jω⋅ c

=e

∆ρ

−4 jπ⋅ λ

(2.57)

has appeared in Equation (2.56). This cofactor, determined by Equation (2.57), plays a very important role. If the radar generates simple searching signals, for example, S(t) 1, Ψ(t) 0, ω = ω0, and ∆ω = 0, then on the basis of Equation (2.56) we can write

( )

R∆enρ, ∆x t , τ = m0

(

∫∫ S (ρ − 0.5∆ρ, x − 0.5∆x) )

× S ρ + 0.5∆ ∆ρ, x + 0.5 ∆x ⋅ e

−2 jω 0

∆ρ c

.

(2.58)

dρ dX

The correlation function given by Equation (2.58) is the correlation function of the slow with the continuous searching signal. In other words, this is the correlation function of the fluctuations caused by the moving radar, antenna scanning, and displacements of scatterers — i.e., the space fluctuations.

Copyright 2005 by CRC Press

1598_book.fm Page 48 Monday, October 11, 2004 1:57 PM

48

Signal and Image Processing in Navigational Systems

2.3.3

The Correlation Function with the Pulsed Searching Signal

Let us assume that the radar generates a sequence of identical, coherent pulsed searching signals with the duration τp << Tp, where Tp is the period of recurrence: ∞

( )

U t, ω =

∑ P (t − kT ) ⋅ e

(

− jω 0 t − kTp

p

),

(2.59)

k=0

where P(t) = Π(t) · e–jΨ(t);

(2.60)

Π(t) is the envelope of the high-frequency pulse (video-signal); Ψ(t) is the phase modulation law within the limits of the pulsed searching signal duration, which is not taken into consideration by the term ω0. For example, Ψ(t) is caused by frequency changes within the pulsed searching signal duration.43,44 To determine the correlation function of target return signal fluctuations, it is necessary to substitute Equation (2.59) in Equation (2.53). At first, we must define the product of two pulse sequences shifted with respect to each other:

(

)

(

(

)

U t , τ , ∆ω ⋅ U * t , τ , ∆ω = U t −

(

2ρ c

×U* t−

− 0.5τ1 , ω 0 − 0.5∆ω 2ρ c

)

+ 0.5τ1 , ω 0 + 0.5 ∆ω

)

,

(2.61)

where τ1 = τ – 2∆ρ c . Thus, |τ| = |t2 – t1| ∈[0,∞). Taking into consideration that τp << Tp and denoting the index of summing the signal U(t, τ, ∆ω) by k and the index of summation of the signal U*(t, τ, ∆ω) by m and assuming n = m – k, we can write

(

)

(

)

U t , τ , ∆ω ⋅ U * t , τ , ∆ω = e

∆ρ 2ρ   − j 2ω 0 ⋅ + ∆ω  t −    c c  

∞

×

∞

∑ ∑ P (t −

2ρ c

− 0.5τ1 + nTp − mTp

m= 0 n= 0

(

× P* t−

2ρ c

)

)

iω τ − nT + 0.5τ1 − nTp − mTp ⋅ e 0 ( p )

(2.62) In the case of coherent pulsed searching signals, we can assume that e–jω0nTp = 1. For symmetry, we can write the expression for the double summation in Equation (2.62) in the following form: Copyright 2005 by CRC Press

1598_book.fm Page 49 Monday, October 11, 2004 1:57 PM

Correlation Function of Target Return Signal Fluctuations ∞

∞

∑ ∑ P t − m= 0 n= 0

2ρ c

(

)

− 0.5 τ1 − nTp − mTp  ⋅ P * t −  

2ρ c

49

(

)

+ 0.5 τ1 − nTp − mTp   (2.63)

Here |τ1 – nTp| ≤ Tp. The function given by Equation (2.62) is periodic with respect to two variables: τ1 – nTp, where nTp is the relative shift between two pulsed searching signal sequences, and t – mTp. Summing with respect to the index, m can be omitted. Then, substituting Equation (2.62) in Equation (2.53), we can write R∆ρ,∆x (t , τ , ∆ω ) = R∆enρ ,∆x (t , τ , ∆ω ) ⋅ e jω 0τ

(2.64)

where

(

)

R∆enρ, ∆x t , τ , ∆ω = m0

∫∫ P t −

× P * t − 

2ρ c

2ρ c

(

)

− 0.5 τ − nTp − ∆ρ  

(

)

+ 0.5 τ − nTp − ∆ρ  

(

× S ρ − 0.5 ∆ρ, x − 0.5 ∆x

(

(2.65)

) )

× S ρ + 0.5 ∆ρ, x + 0.5 ∆x ⋅ e

∆ρ 2ρ   − j  2 ω 0 ⋅ + ∆ω  t −    c c  

dρ dX

is the envelope of the high-frequency correlation function R∆ρ,∆x (t, τ, ∆ω). The high-frequency correlation function R∆ρ,∆x (t, τ, ∆ω) given by Equation (2.64) defines both fluctuations in the radar range (the intraperiod fluctuations) and fluctuations from period to period (the interperiod fluctuations). In the case of the nonstationary target return signal, R∆ρ,∆x (t, τ, ∆ω) depends on time. So long as the differentials ∆ρ and ∆x are thought of as independent, i.e., independent of the parameters τ and t, the envelope R∆enρ,∆x (t, τ, ∆ω) is a periodic function of the parameters τ and t. If we assume that P(t) 1 and ∆ω = 0, then the envelope R∆enρ,∆x (t, τ, ∆ω) transforms to the envelope of the correlation function determined by Equation (2.58). Thus, the envelope R∆enρ,∆x (t, τ, ∆ω) can be used with both pulsed searching signals and with continuous searching signals in navigational systems.45,46

2.3.4

The Average Correlation Function

In the case of the periodic searching signal, the correlation function of target return signal fluctuations too, as was noted previously, is a periodic function with respect to time. To determine the average (with respect to time) correlation function, we must average the time correlation function within the limits of the period of the pulsed searching signal:

Copyright 2005 by CRC Press

1598_book.fm Page 50 Monday, October 11, 2004 1:57 PM

50

Signal and Image Processing in Navigational Systems

0.5Tp

( )

1 R t, τ = Tp

∫ R (t, τ) dt .

(2.66)

− 0.5Tp

To define the average correlation function given by Equation (2.66), we sometimes use the following technique. Represent the searching signal given by Equation (2.59) in the Fourier series form:

()

∞

U t =

∑N ⋅e

(

) ,

− j ω 0 + kΩ p t

k

(2.67)

k=0

where 1 Nk = Tp

0.5Tp

( ∫ U (t ) ⋅ e

) dt

j ω 0 + kΩ p t

(2.68)

− 0.5Tp

and Ωp is the frequency of the pulsed searching signal. Substituting Equation (2.67) in Equation (2.53) under the condition ∆ω = 0, we can write

( )

∞

R∆ρ, ∆x t , τ = m0

∞

∑∑

Nk Nn* ⋅ e

(

)

0.5 j ω 0 + n + k Ω p  τ

k = 0 n= 0

(

⋅

∫∫

e

2ρ 2 ∆ρ   − j  k − n Ω p  t −  + ω 0 + 0.5 n + k Ω p  ⋅   c  c  

(

)

(

) (

)

)

× S ρ − 0.5 ∆ρ, x − 0.5 ∆x ⋅ S ρ + 0.5 ∆ρ, x + 0.5 ∆x dρ dX . (2.69) The correlation function determined by Equation (2.69) is equivalent to the correlation function given by Equation (2.56), but the variables t and τ are separable. This fact allows us to average the correlation function given by Equation (2.66) in a simpler form. Integrating with respect to the variable t and taking into consideration the fact that the terms in Equation (2.68) differ from zero only under the condition k = n, we can write

( )

R∆ρ, ∆x t , τ =

∞

∑N

n

n= 0

2

(

)

⋅ Rcon τ , ω 0 + nΩ p ,

(2.70)

where Rcon(τ, ω0 + nΩp) is the correlation function in the case of the continuous nonmodulated searching signal given by Equation (2.58), in which the frequency ω0 is replaced with ω0 + nΩp. Reference to Equation (2.70) shows we can conclude that the average power spectral density with respect to time Copyright 2005 by CRC Press

1598_book.fm Page 51 Monday, October 11, 2004 1:57 PM

Correlation Function of Target Return Signal Fluctuations

51

of the target return signal fluctuations for any kind of modulation law is defined by the sum of the power spectral densities Scon(ω) at frequencies ω0 + nΩp:

( )

S ω, t =

2.4 2.4.1

∞

∑N

n

n= 0

2

(

)

⋅ Scon ω − ω 0 − nΩ p .

(2.71)

The Correlation Function under Scanning of the ThreeDimensional (Space) Target General Statements

Using Equation (2.64), we can find the correlation function of the target return signal fluctuations in the cases of continuous and pulsed searching signals. The amplitude of the target return signal from an arbitrary scatterer can be determined using the main radar equation S(ρ) =

Ps G0 λ g(ϕ , ψ )q(ξ , ζ) 8 π 3 ρ2

,

(2.72)

where Ps is the power of the searching signal; λ is the wavelength; ϕ and ψ are the angles defining the position of the scatterer relative to the radar antenna directional diagram axis in the main planes (see Figure 2.5); G0 is the radar antenna amplifier coefficient; g(ϕ, ψ) is the normalized two-dimensional directional diagram under the condition g(0,0) = 1; ξ and ζ are the angles defining the position of the scatterer in space relative to the radar antenna polarization plane and direction of beam, as shown in Figure 2.6, where it is assumed that scatterers possess axis symmetry and that two angles exactly define the orientation of scatterers; q(ξ, ζ) is the function representing the dependence of the target return signal amplitude on the orientation of the scatterer in space and q2(ξ, ζ) is the effective scattering area with fixed values of the angles ξ and ζ; and ρ is the radar range. The volume element using the coordinates ρ, ϕ, and ψ is equal to ρ2 dρ dϕ dψ and the corresponding Jacobian is equal to ρ2. The angles ξ and ζ belong to the spherical coordinate system. Under integration over all orientations of scatterers, the surface element can be written in the form: sin ζ dξ dζ. If the directional diagrams are different under conditions of transmission and reception, we can write g(ϕ, ψ) = gt(ϕ, ψ) · gr(ϕ, ψ),

Copyright 2005 by CRC Press

(2.73)

1598_book.fm Page 52 Monday, October 11, 2004 1:57 PM

52

Signal and Image Processing in Navigational Systems ψ

ρ ∆h

ϕ

∆ν

FIGURE 2.5 The coordinate system using the variables ϕ and ψ.

Z

Electromagnetic Wave ξ

X

ζ

Y FIGURE 2.6 The coordinate system using the variables ξ and ζ.

where gt(ϕ, ψ) and gr(ϕ, ψ) are the directional diagrams under conditions of transmission and reception by voltage for the general coordinate system ϕ and ψ, respectively. Thus, the amplitude of the target return signal is the function of five variables: ρ, ϕ, ψ, ξ, and ζ. Substituting Equation (2.72) in Equation (2.64) and using the condition ∆ω = 0 (which is equivalent to the fact that a rotation of scatterers ∆ξ and ∆ζ is independent of their positions within the directional diagram, in other words, independent of the coordinates ρ, ϕ, and ψ, and

Copyright 2005 by CRC Press

1598_book.fm Page 53 Monday, October 11, 2004 1:57 PM

Correlation Function of Target Return Signal Fluctuations

53

that the angle differentials ∆ϕ and ∆ψ are independent of the orientations of scatterers, in other words, independent of the coordinates ξ and ζ, and using the asymmetric form, we obtain47,48 R∆enρ,∆ϕ ,∆ψ ,∆ξ ,∆ζ (t , t + τ) = p(t) ⋅ R∆enρ ,∆ϕ ,∆ψ ,∆ξ ,∆ζ (t , t + τ) ,

(2.74)

where

(

)

(

)

(

R∆enρ, ∆ϕ , ∆ψ , ∆ξ , ∆ζ t , t + τ = R∆ρ, ∆ϕ , ∆ψ t , t + τ ⋅ R∆ξ , ∆ζ ∆ξ , ∆ζ

)

(2.75)

is the total normalized correlation function of the fluctuations;

(

∞

)

R∆ρ, ∆ϕ , ∆ψ t , t + τ = N ⋅

∑ ∫∫∫ P (t − ) ⋅ P * (t − 2ρ c

2ρ c

+ τ − nTp −

2 ∆ρ c

n= 0

×

(

) ( (ρ + ∆ρ)

g ϕ , ψ ⋅ g ϕ + ∆ϕ , ψ + ∆ψ 2

) ⋅e

) (2.76)

− jω 0 ⋅

2 ∆ρ c

dρ dϕ dψ

and

(

)

R∆ξ , ∆ζ ∆ξ , ∆ζ = N

∫∫ q ( ξ, ζ) ⋅ q ( ξ + ∆ξ, ζ + ∆ζ) sin ζ dξ dζζ

(2.77)

are the normalized correlation functions: R∆ρ,∆ϕ,∆ω(t, t + τ) is the normalized correlation function of the fluctuations caused by the moving radar, displacements of scatterers, and radar antenna scanning; R∆ξ,∆ζ(∆ξ, ∆ζ) is the normalized correlation function of the fluctuations caused by rotation of scatterers and radar antenna scanning; N is the normalized coefficient [see Equation (2.32)];

()

p t =

m0 PSG02 λ 2 ⋅ 64 π 3

∫

(

Π2 t − ρ

2

2ρ c

) dρ

∫∫ g (ϕ, ψ ) dϕ dψ ∫∫ q ( ξ, ζ) sin ζ dξ dζ 2

2

(2.78) is the received target return signal power at the instant of time t.

2.4.2

The Correlation Function with the Pulsed Searching Signal

The total normalized correlation function of the fluctuations given by Equation (2.75) is the product of two normalized correlation functions. The first function defines the slow fluctuations caused by the moving radar, Copyright 2005 by CRC Press

1598_book.fm Page 54 Monday, October 11, 2004 1:57 PM

54

Signal and Image Processing in Navigational Systems

displacements of scatterers (∆ρ), and antenna scanning (∆ϕ and ∆ψ) and the rapid fluctuations caused by propagation of the searching signal in space (τ). The second function defines the slow fluctuations caused by rotation of scatterers and the radar antenna polarization plane (∆ϕ and ∆ζ). Reference to Equation (2.76) shows that with moving radar (∆ρ ≠ 0), there are three sources giving rise to target return signal fluctuations: phase changes in elementary signals –∆ρ in the exponent of the exponential function; amplitude changes in elementary signals caused by the duration and shape of the searching signal envelope –∆ρ is the argument of the function P(t); and amplitude changes in elementary signals caused by searching signal attenuation as a function of distance between the radar and scatterer –∆ρ in the sum ρ + ∆ρ. Fluctuations caused by phase changes in elementary signals are the most rapid fluctuations. Amplitude changes in elementary signals caused by the searching signal envelope give rise to the slow fluctuations. Attenuation of the searching signal caused by increasing the distance between the radar and the scatterer gives rise to slower fluctuations.49,50 Because of this, we can neglect the differential ∆ρ in the sum ρ + ∆ρ. This statement will be rigorously proved in Section 2.4.4. To define the position of scatterers with respect to the radar and to the direction of moving radar, we introduce the spherical coordinate system that is functionally related to the radar: the distance ρ, the azimuth β, and the aspect angle γ (see Figure 2.7). The direction of moving radar in the spherical coordinate system is characterized by the angles β and γ, and the position of the directional diagram axis is characterized by the angles β0 and γ0. We must orient the coordinate system (ϕ, ψ) functionally related to the directional diagram axis in such a manner that the plane ψ(ϕ = 0) can be matched with the plane γ. Then we can write ϕ = (β – β0) cos γ

and

ψ = γ – γ0 .

(2.79)

When scatterers are stationary, the differentials ∆ϕ and ∆ψ of the angle coordinates ϕ and ψ are caused, in the general case, by two sources: antenna scanning (∆ϕsc and ∆ψsc) and moving radar (∆ϕrm and ∆ψrm). The differentials ∆ϕsc and ∆ψsc are defined by the shifts ∆β0 and ∆γ0 of the directional diagram axis during antenna scanning. Reference to Equation (2.79) shows that ∆ϕsc = ϕ(β0) – ϕ(β0 + ∆β0) = ∆β0 cos γ

(2.80)

∆ψsc = ψ(γ0) – ψ(γ0 + ∆γ0) = ∆γ0.

(2.81)

and

The differentials ∆ϕrm and ∆ψrm are caused by changes in the azimuth of scatterers when the radar moves over a distance ∆ (see Figure 2.8):

Copyright 2005 by CRC Press

1598_book.fm Page 55 Monday, October 11, 2004 1:57 PM

Correlation Function of Target Return Signal Fluctuations

ψ

55

ϕ S (ρ)

γ

γ0 β0

β

ρ

0

FIGURE 2.7 The coordinate system using the variables ρ, β, and γ for the three-dimensional (space) target.

∆ ∆ρ ρ

v 0

θ

γ β

FIGURE 2.8 Radial displacement of scatterers with moving radar.

Copyright 2005 by CRC Press

1598_book.fm Page 56 Monday, October 11, 2004 1:57 PM

56

Signal and Image Processing in Navigational Systems ∆ ⋅ sin β ρ

(2.82)

∆ ⋅ cos β sin γ . ρ

(2.83)

∆ϕ rm = and ∆ψ rm =

In the majority of cases, with high values of ρ, the differentials ∆ϕrm and ∆ψrm are very small in comparison with the directional diagram width.51,52 Consequently, we can neglect amplitude changes in elementary signals that are caused by the differentials ∆ϕrm and ∆ψrm. An exception to this rule is the neighboring effective area of the directional diagram, where the differentials ∆ϕrm and ∆ψrm given by Equation (2.82) and Equation (2.83), respectively, must be taken into consideration. This is very important, especially in the use of laser antenna, for which the neighboring effective area of the directional diagram is large. As can be seen from Equation (2.80)–Equation (2.83), the differentials ∆ϕ and ∆ψ differ for various scatterers because they depend on the coordinates ρ, β, and γ. Let us assume that the range of changes for the coordinates ρ, β, and γ within the resolution area forming the resulting target return signal is not so large. Then the coordinates ρ, β, and γ can be changed to the coordinates ρ*, β0, and γ0, respectively, where ρ* is the distance between the center of the pulse volume and radar. Also, we can assume that the differentials ∆ϕ and ∆ψ are the same for all scatterers. In other words, we can write the following formulae ∆ϕsc = ∆β0 cos γ0 ;

(2.84)

∆ψsc = ∆γ0 ;

(2.85)

∆ ⋅ sin β 0 ; ρ*

(2.86)

∆ ⋅ cos β 0 sin γ 0 ρ*

(2.87)

∆ϕ rm =

∆ψ rm =

instead of Equation (2.80)–Equation (2.83). Thus, using Equation (2.84)–Equation (2.87), we can write ∆ϕ = ∆ϕ sc + ∆ϕ rm = ∆β 0 cos γ 0 +

Copyright 2005 by CRC Press

∆ ⋅ sin β 0 ρ*

(2.88)

1598_book.fm Page 57 Monday, October 11, 2004 1:57 PM

Correlation Function of Target Return Signal Fluctuations

57

and ∆ψ = ∆ψ sc + ∆ψ rm = ∆γ 0 +

∆ ⋅ cos β 0 sin γ 0 . ρ*

(2.89)

This representation of the differentials ∆ϕ and ∆ψ makes determination more simple, and allows us to define the normalized correlation function R∆ρ,∆ϕ,∆ψ (t, t + τ) given by Equation (2.76) as a function of the shifts ∆β0 and ∆γ0 of the directional diagram axis and the displacement ∆ of the radar. When the radar moves over a distance ∆ the radial displacements of scatterers can be determined as follows (see Figure 2.8): ∆ρ = ∆ cos θ −

∆ 2 ⋅ sin 2 θ 0 , 2ρ

(2.90)

where θ is the angle between the velocity vector and the direction of scatterers, which is functionally related to the angles β and γ by cos θ = cos β · cos γ.

(2.91)

The second term in Equation (2.90) is infinitesimal and we can neglect it. Using Equation (2.79) and Equation (2.91), we can write

(

∆ρ = ∆ ⋅ cos β 0 +

ϕ cos γ

) ⋅ cos(γ

0

+ ψ ) = ∆ρ0 + ∆ρϕ ,ψ ,

(2.92)

where ∆ρ0 = ∆ cos β0 cos γ0 is the radial shift along the directional diagram axis; ∆ρϕ,ψ = ∆ × f(ϕ, ψ) is the deviation of radial shifts for various scatterers within the directional diagram. Thus, the displacement ∆ρ is a function of the angles ϕ and ψ. In other words, the displacement ∆ρ depends on the position of scatterers within the resolution area. This function should be jω ∆ρ taken into account in the exponent 0c of the exponential function in Equation (2.76), which defines phase changes in elementary signals with displacements of scatterers because variations in the value of ∆ρ can be compared with the wavelength λ. However, we can neglect the dependence between the parameters ∆ρ, ϕ, and ψ in the argument of the pulsed function P(t) because variations in the value of ∆ρ are small in comparison with the length of the function P(t) and we can assume that ∆ρ = ∆ρ0. This assumption allows us to divide the triple integral in Equation (2.76) between the double integral with limits of integration over the variables ϕ and ψ and the simple integral with limits of integration over the variable ρ. As usual, the interval of changes in distance within the duration of the pulsed searching signal is not so large. Hence, we can replace the variable ρ in the denominator of Equation (2.76) with the mean of the variable ρ* and factor

Copyright 2005 by CRC Press

1598_book.fm Page 58 Monday, October 11, 2004 1:57 PM

58

Signal and Image Processing in Navigational Systems

the variable ρ* outside the integral sign. We introduce a variable z = t – Then R∆ρ,∆ϕ,∆ψ(t, t + τ) = Rg(∆, ∆β0, ∆γ0) · Rp(τ, ∆),

2ρ c

.

(2.93)

where

(

)

Rg ∆, ∆β 0 , ∆γ 0 = N

∫∫

(

) (

)

g ϕ , ψ ⋅ g ϕ + ∆ϕ , ψ + ∆ψ ⋅ e

−

(

4 jπ∆ρ ϕ , ψ λ

)

dϕ dψ (2.94)

and

(

)

∞

Rp τ , ∆ = N

∑ ∫ P ( z) ⋅ P * ( z + τ − nT − ) dz . p

2 ∆ρ0 c

(2.95)

n= 0

Thus, the normalized correlation function of the target return signal fluctuations in navigational systems caused by the moving radar and antenna scanning is defined by the product of two normalized correlation functions with the pulsed searching signal. The first normalized correlation function defines the slow fluctuations caused by the different Doppler shifts in the frequency of elementary signals within the resolution area of pulse volume with moving radar, amplitude changes in elementary signals due to antenna scanning, and variation of the aspect angle during radar motion; these are the interperiod fluctuations. The second normalized correlation function is a periodic function of the variable τ and defines the rapid fluctuations caused by propagation of the pulsed searching signal in space with respect to simultaneous radar motion; these are the intraperiod fluctuations. For the considered approximation, the second normalized correlation function is independent of time or the radar range.

2.4.3

The Target Return Signal Power with the Pulsed Searching Signal

The power p(t) of the target return signal is determined by Equation (2.78), for which St =

∫∫ q (ξ, ζ) ⋅ sin ζ dξ dζ 2

(2.96)

is the effective scattering area of a scatterer, which is averaged over all possible positions of the scatterer in space. Thus, S° = m0St is the specific effective scattering area. In other words, this is the effective scattering area

Copyright 2005 by CRC Press

1598_book.fm Page 59 Monday, October 11, 2004 1:57 PM

Correlation Function of Target Return Signal Fluctuations

59

of the pulse unit volume filled by a scatterer. In the determination of the double integral, we can assume that g(ϕ, ψ) = gh(ϕ) · gv(ψ),

(2.97)

where gh(ϕ) and gv(ψ) are the normalized radar antenna directional diagrams by power for two mutually perpendicular (main) planes, the conditionally horizontal plane and the vertical plane. Then

∫∫ g (ϕ, ψ )dϕ dψ = ∫ g (ϕ)dϕ ⋅ ∫ g (ψ ) = ∆ 2

2 h

(2)

2 v

h

⋅ ∆(v2 ) ,

(2.98)

where ∆(h2 ) and ∆(v2 ) are the effective widths of the squares of the directional diagram by power for the horizontal and vertical planes. Usually, the radar antenna is characterized by the effective widths ∆h and ∆v of the first order (not of the second order) of the directional diagram by power: ∆ = ∫ g(x)dx.53–55 The widths ∆h and ∆v of the first and second orders are functionally related: ∆(h2 ) = k ah ∆ h

and ∆(v2 ) = k av ∆ v ,

(2.99)

where k ha and k va are coefficients of the shape of the directional diagram, respectively, which differ from unity. Henceforth, we will consider two kinds of directional diagram: the Gaussian and sinc2 models. The two-dimensional Gaussian directional diagram model takes the following form:

(

)

g ϕ, ψ = e

 2  ψ2  ϕ − π +  (2) 2 ( )  ∆ ∆ v  h

( ) ( )

= gh ϕ ⋅ gv ψ .

(2.100)

It defines exactly the shape of the major lobe, and is also very convenient mathematically. Unlike the actual directional diagrams, the two-dimensional Gaussian directional diagram model does not have side-lobes. If the twodimensional Gaussian directional diagram model has axis symmetry ∆h = ∆v = ∆a, then

(

)

− π⋅

g ϕ, ψ = e

ϕ2 + ψ 2 2 ∆a

()

2 −π ⋅ θ

=e

()

2 ∆a

()

=g θ ,

(2.101)

where θ2 = ϕ2+ ψ2. For the two-dimensional Gaussian directional diagram model, the following equality k ah = k av = k a = 2 −0.5

Copyright 2005 by CRC Press

(2.102)

1598_book.fm Page 60 Monday, October 11, 2004 1:57 PM

60

Signal and Image Processing in Navigational Systems

is true. The two-dimensional sinc2 directional diagram model takes the following form: g(ϕ , ψ ) = g h (ϕ) ⋅ g v ( ψ ) = sinc 2

πϕ πψ ⋅ sinc 2 ∆h ∆v

(2.103)

and corresponds approximately to uniform radiation of a rectangle aperture and has very large side-lobes. This model of the directional diagram does not possess axis symmetry even if the condition ∆h = ∆v is true. In the case of the two-dimensional sinc2 directional diagram model, the equality k ah = k av = k a = 23 is true. The two-dimensional Gaussian and sinc2 directional diagram models given by Equation (2.100) and Equation (2.103), respectively, can be considered, in principle, as two extreme cases limiting a set of directional diagram models with various levels of side radiation. The coefficient ka is approximately the same for various directional diagram models in spite of differences in shapes. In other words, in the determination of target return signal power, the shape of the directional diagram model plays only a minor role. In the case of the rectangle directional diagram model, we can write ka = 1. But the rectangle directional diagram model is not realizable in practice and, for this reason, is not studied further. When the pulsed searching signal has a short duration, the third integral can be written in the following form:

∫

(

Π2 t − ρ

2

2ρ c

) dp ≈

∫

c c c ⋅ Π 2 (t) dt = τ (p2 ) ⋅ 2 = k p τ p ⋅ 2 , 2ρ*2 2ρ* 2ρ*

(2.104)

where τ (p2 ) = kpτp is the effective duration of the square of the envelope of the pulsed searching signal; kp is the coefficient of the shape of the pulsed searching signal. In the case of the square waveform pulsed searching signal, the equality kp = 1 is true. In the case of the Gaussian pulsed searching signal, we can write Π( t ) = e

2 − π⋅ t

(2) τp

,

(2.105)

where τp is the effective duration of the pulsed searching signal and kp = 2–0.5. Substituting Equation (2.96), Equation (2.98), Equation (2.99), and Equation (2.104) in Equation (2.78), we can write

()

p t =

Copyright 2005 by CRC Press

PSG02 λ 2 S°∆ h ∆ v cτ p kah kav kp 128π 3 ρ*2

,

(2.106)

1598_book.fm Page 61 Monday, October 11, 2004 1:57 PM

Correlation Function of Target Return Signal Fluctuations

61

where ρ* = 0.5ct. Comparing Equation (2.106) with the radar equation for the point target with effective scattering area equal to St p(t) =

PSG02 λ2St , 64 π 3ρ 4

(2.107)

one can see that the total target return signal power from a three-dimensional target can be defined using the radar equation for a single target, if the effective scattering area has the form:56,57 Stspace = S° · Q, where Q = 0.5 ⋅ cτ (p2 ) ∆(h2 ) ∆(v2 )ρ*2 = 0.5 ⋅ k p k ah k av cτ p ∆ h ∆ v ρ*2

(2.108)

is the pulse or resolution volume in space. Thus, under scanning of the threedimensional (space) target, the target return signal power at the receiver or detector input in navigational systems is inversely proportional to ρ2, but not to ρ4, as in the case of the point target. This can be explained by the fact that with an increase in distance between the radar and the three-dimensional (space) target, the pulse volume and effective scattering area increase proportionally to ρ2.

2.4.4

The Correlation Function and Power of the Target Return Signal with the Simple Harmonic Searching Signal

In the considered case, Equation (2.65) is true if the condition P(t) 1

(2.109)

is satisfied. Formally, in the following formulae, the distance ρ in integrands is not equal to ρ* = const because the differentials ∆ϕ and ∆ψ in Equation (2.88) and Equation (2.89) depend on the distance ρ, which is a parameter of the functions given by Equation (2.76) and Equation (2.78). For this reason, it is necessary to integrate over the entire area occupied by scatterers but, as before, we can neglect the differential ∆ρ in the sum ρ + ∆ρ or, with rigorous analysis, we can assume that ∆ρ = ∆ρ0 = ∆ cos β0 cos γ0. On the basis of the normalized correlation function of the target return signal fluctuations given by Equation (2.75), we can, as before, isolate the normalized correlation function R∆ξ,∆ζ(∆ξ, ∆ζ) determined by Equation (2.77) in the form of an individual cofactor. The remaining continuous normalized correlation function R∆con ρ , ∆ϕ , ∆ψ (t, t + τ) takes the following form:

(

)

R∆con ρ, ∆ϕ , ∆ψ t , t + τ = N

∫∫∫

(

) ( (ρ + ∆ρ )

g ϕ , ψ ⋅ g ϕ + ∆ϕ , ψ + ∆ψ 2

) ⋅e

−2 jω 0 ⋅

∆ρ c

dρ dϕ dψ .

0

(2.110)

Copyright 2005 by CRC Press

1598_book.fm Page 62 Monday, October 11, 2004 1:57 PM

62

Signal and Image Processing in Navigational Systems

In the long range of the radar antenna directional diagram, which is of prime interest to us, as a rule, the differentials ∆ϕ and ∆ψ are independent of the distance ρ.58 Then,

(

)

(

) ( )

R∆con ρ, ∆ϕ , ∆ψ t , t + τ = Rg ∆, ∆β 0 , ∆γ 0 ⋅ Rρ ∆ ,

(2.111)

where Rg(∆, ∆β0, ∆γ0) is determined by Equation (2.94) under the condition ∆ϕrm = ∆ψrm = 0 and dρ

∫ (ρ + ∆ρ ) R ( ∆) = dρ ∫ρ

2

0

ρ

.

(2.112)

2

If the parameter ρ1 is the minimum distance to receive the target return signal, then integrating the formula in Equation (2.112) within the limits of the interval [ρ1, ∞), we can write Rρ ( ∆) =

1 ∆ρ 0 ρ1

1+

=

1 1+

∆ cos β0 cos γ 0 ρ1

.

(2.113)

The correlation interval of the corresponding fluctuations has the following form ∞

∆ ρ =

∫ R ( ∆) d ( ∆) → ∞ , ρ

(2.114)

−∞

which indicates a very small bandwidth for the power spectral density of these fluctuations, allowing us to neglect them. Thus, in the long range of the directional diagram, the normalized correlation function of the fluctuations with a continuous nonmodulated searching signal has the same shape as the normalized correlation function of the fluctuations given by Equation (2.94), which are caused by the moving radar and antenna scanning with the pulsed searching signal. The target return signal power from the three-dimensional (space) target with the continuous searching signal can be determined by assuming in Equation (2.78) that the condition given by Equation (2.109) is true within the limits of the interval [ρ1, ρ2] occupied by scatterers. Instead of Equation (2.104) we can write ρ2

∫ρ

dρ 2

ρ1

Copyright 2005 by CRC Press

= ρ1−1 − ρ−21 .

(2.115)

1598_book.fm Page 63 Monday, October 11, 2004 1:57 PM

Correlation Function of Target Return Signal Fluctuations

63

Under the condition ρ2 >> ρ1, we can neglect the second term in Equation (2.115). Then the target return signal power with the simple harmonic searching signal has the following form:

()

p t =

PSG02 λ 2S°∆ h ∆ v kah kav . 64 π 3 ρ1

(2.116)

Reference to Equation (2.116) shows that the target return signal power with the simple harmonic searching signal decreases proportionally to the first order of the distance ρ1 between the radar and front border of the threedimensional (space) target. The region around the point ρ1 = 0 is not considered. As ρ1 → 0, the target return signal power tends to approach ∞, i.e., p(t) → ∞. This is called the “effect of dazzley.” Comparing Equation (2.107) and Equation (2.116), one can see that the effective scattering area of the three-dimensional (space) target with the continuous pulsed searching signal has the following form:

(2) 2 Stspace = S°∆ h ∆ (v )ρ13 = S°∆ h ∆ v kah kav ρ13 .

2.5 2.5.1

(2.117)

The Correlation Function in Angle Scanning of the TwoDimensional (Surface) Target General Statements

In scanning of the two-dimensional (surface) target, the radar range is not an independent coordinate.59,60 The parameter ρ is a function of the altitude h and aspect angle γ (see Figure 2.9) and is determined as follows: ρ=

h h , = sin γ sin( γ 0 + ψ )

(2.118)

where γ0 is the aspect angle of the radar antenna directional diagram axis. The element of the searched two-dimensional (surface) target can be determined using the coordinates ϕ and ψ as follows: dX = J dϕ dψ =

Copyright 2005 by CRC Press

h2 dϕ dψ . sin ( γ 0 + ψ ) 3

(2.119)

1598_book.fm Page 64 Monday, October 11, 2004 1:57 PM

64

Signal and Image Processing in Navigational Systems V ε0

∆v ψ∗ h

ψ ρ0

∆p

γ∗

γ

ρ

ρ∗

γ0

c τp 2 cosγ∗ FIGURE 2.9 The coordinate system under scanning of the three-dimensional (space) target.

Furthermore, it is necessary to take into consideration the dependence of the effective scattering area S° of the two-dimensional (surface) target as a function of the aspect angle γ and azimuth β. Hence, we can write m0q2(ξ, η) = S°(β, γ) = S°(ϕ + β0, ψ + γ0)

(2.120)

instead of the function q2(ξ, η). The function S°(β, γ) can be considered as the back-scattering diagram of the two-dimensional (surface) target. These representations are equivalent. As a rule, the surface of the two-dimensional target is assumed to be weakly nonisotropic. The reflectance of the surface of the two-dimensional target is a very slow function of the azimuth.61,62 Because of this, we can neglect this function within the directional diagram. Then we can replace the variable β = β0 + ϕ with the variable β0 in Equation (2.120). The majority of real surfaces of two-dimensional targets satisfy this condition. The target return signal amplitude under this condition takes the following form:

()

S ρ =

(

PS G0 λ g ϕ , ψ

)

(

)

(

S° β 0 , ψ + γ 0 sin 2 ψ + γ 0 8 π h 3

2

)

(2.121)

instead of Equation (2.72). Using Equation (2.118)–Equation (2.121), assuming ∆ω = 0, and neglecting the displacement ∆ρ in the function sin(ψ + γ0) [which is equivalent to neglecting the component ∆ρ in the sum ρ + ∆ρ in Equation (2.76)], on the basis of Equation (2.65) we can write

Copyright 2005 by CRC Press

1598_book.fm Page 65 Monday, October 11, 2004 1:57 PM

Correlation Function of Target Return Signal Fluctuations ∞

R

en ∆ρ, ∆ϕ , ∆ψ

(t, τ ) = p ∑ ∫∫ P t −

( )

2ρ ψ

0

c

(

( )

2ρ ψ c

)

− 0.5 τ − nTp +

n= 0

× P * t − 

65

(

)

+ 0.5 τ − nTp −

(

∆ρ c

∆ρ c

 

 

) (

× g ϕ − 0.5 ∆ϕ , ψ − 0.5∆ψ ⋅ g ϕ + 0.5∆ϕ , ψ + 0.5∆ψ ×e

−2 jω 0 ⋅

∆ρ c

(

) (

)

)

S° β0 , ψ + γ 0 sin ψ + γ 0 dϕ dψ (2.122)

where p0 =

PSG02 λ 2 . 64 π 3 h2

(2.123)

Assuming that ∆ρ = ∆ϕ = ∆ψ = τ = 0 in Equation (2.122) and omitting the summation sign, we can determine the target return signal power from the two-dimensional (surface) target at the instant of time t in the following form:

()

p t = p0

2.5.2

∫∫ Π

2

t − 2 ρ( ψ )  ⋅ g 2 ϕ , ψ ⋅ S° β , ψ + γ sin ψ + γ dϕ dψ . 0 0 0 c    (2.124)

(

) (

) (

)

The Correlation Function with the Pulsed Searching Signal

Suppose the angle between the direction of moving radar and the horizon is equal to ε0 (see Figure 2.9). One can see that the differentials of coordinates can be determined as follows: ∆ϕ = ∆ϕ sc + ∆ϕ rm = ∆β 0 cos γ * +

∆ψ = ∆ψ sc + ∆ψ rm = ∆γ 0 +

[

(

∆ρ = ∆ cos ε 0 cos β 0 +

∆ ⋅ cos ε 0 sin β 0 ; ρ*

(2.125)

∆ (cos ε 0 cos β 0 sin γ * + sin ε 0 cos γ * ) ; ρ* (2.126)

ϕ cos γ *

) cos(ψ + γ ) − sin ε 0

0

]

sin(ψ + γ 0 ) , (2.127)

where γ* = γ0 + ψ* is the aspect angle of the middle of the two-dimensional (surface) target resolution element; and ε0 > 0 if the altitude is increased with moving radar. Here, we can assume that the differentials ∆ϕ and ∆ψ are Copyright 2005 by CRC Press

1598_book.fm Page 66 Monday, October 11, 2004 1:57 PM

66

Signal and Image Processing in Navigational Systems

approximately the same for all scatterers within the resolution element, as in Section 2.4.2. On the basis of the preceding observations, and taking Equation (2.95) into consideration, the differential ∆ρ (displacement), which is an argument of the function P(t), can be thought of as identical for all scatterers and has the following form: ∆ρ* = ∆(cos ε0 cos β0 cos γ* – sin ε0 sin γ*).

(2.128)

We cannot neglect the dependence of the differential ∆ρ on the coordinates −

2 jω 0 ∆ρ

ϕ and ψ of scatterers in the exponential function e c in Equation (2.127) because differences in phase changes of various scatterers (differences in the Doppler frequency) are the main source of fluctuations with moving radar, as a rule. Equation (2.122) can be simplified if we assume that the approximate equalities S°(β0, ψ + γ0) ≈ S°(β0, γ*) and sin(ψ + γ*) ≈ sin γ* are true within the two-dimensional (surface) target resolution element. Unlike the case of the three-dimensional (space) target, the normalized correlation function of target return signal fluctuations given by Equation (2.122) cannot be expressed as the product of the normalized correlation functions of the intraperiod and interperiod fluctuations. However, using the fact that the interval of variations of the angle ψ within the duration of the pulsed searching signal is infinitesimal, we can determine this normalized correlation function as the product of two normalized correlation functions with alternative physical meanings. Under scanning of the two-dimensional (surface) target by short-duration pulsed searching signals, so that the angle dimension of the resolution element on the plane γ is satisfied by the condition ∆p << ∆v (see Figure 2.9), variables of the function g(ϕ, ψ) can always be separated. So, we can write g(ϕ, ψ) = g(ϕ, ψ + κ) ≅ gh(ϕ, ψ*) · gv(ψ* + κ) = gh(ϕ, ψ*) · gv(ψ), (2.129) where κ = ψ – ψ* is the angle on the plane ψ, which is determined within the resolution element, and the origin of the angle κ is the mean of the angle ψ*. Thus, the rigorous condition given by Equation (2.127) need not be satisfied. So, we can write ∆ρ(ϕ, ψ) ≅ ∆ρβ(ϕ) + ∆ργ(ψ),

(2.130)

where

[

(

∆ρβ (ϕ) = ∆ cos ε 0 cos β 0 + and

Copyright 2005 by CRC Press

ϕ cos γ *

) cos γ

*

]

− sin ε 0 sin γ * ,

(2.131)

1598_book.fm Page 67 Monday, October 11, 2004 1:57 PM

Correlation Function of Target Return Signal Fluctuations

67

∆ργ(ψ) = –∆(ψ – ψ*) (cos ε0 cos β0 sin γ* + sin ε0 cos γ*).

(2.132)

Using Equation (2.129) and Equation (2.130), on the basis of Equation (2.122) we can write R∆enρ,∆ϕ ,∆ψ (t , τ) = R en ( ∆, ∆β 0 , ∆γ 0 , τ , t) = p(t) ⋅ R en ( ∆, ∆β 0 , ∆γ 0 , τ , t) = = p(t) ⋅ Rβ ( ∆, ∆β 0 ) ⋅ Rγ ( ∆, ∆γ 0 , τ , t)

,

(2.133) where

(

)

(

∫

) (

)

Rβ ∆, ∆β 0 = N gh ϕ − 0.5∆ϕ , ψ * ⋅ gh ϕ + 0.5∆ϕ , ψ * ⋅ e

−2 jω 0 ⋅

( )

∆ρβ ϕ c

dϕ ; (2.134)

(

∞

)

Rγ ∆, ∆γ 0 , τ , t = N

∑ ∫ P t −

( )

2ρ ϕ c

(

n= 0

× P * t − 

( )

2ρ ϕ c

)

− 0.5 τ − nTp +

(

)

+ 0.5 τ − nTp –

(

) (

∆ρ* c

)

()

(

)

∫ (ϕ, ψ ) dϕ ⋅ ∫ Π *

2

 

 

× gv ψ − 0.5 ∆ψ ⋅ gv′ ψ + 0.5 ∆ψ ⋅ e

p t = p0S° β 0 , γ * sin γ *

∆ρ* c

(2.135) −2 jω 0 ⋅

( )

∆ρ ψ c

dψ

t − 2 ρ( ψ )  ⋅ g 2 ψ dψ , (2.136) c    v

( )

and the distance ρ is functionally related to the angle ψ by Equation (2.118). Here, the space normalized correlation function Rβ(∆, ∆β0) defines the slow fluctuations caused by the moving radar with varying elementary signal phase changes ∆ρβ(ϕ) in the azimuth plane and by the rotation ∆β0 of the radar antenna directional diagram axis. The space–time normalized correlation function Rγ(∆, ∆γ0, t, τ) defines the slow fluctuations caused by the moving radar with varying elementary signal phase changes ∆ργ(ψ) in the plane with the aspect angle γ, plane and rotation ∆γ0 of the radar antenna axis, and the rapid fluctuations caused by propagation of the pulsed searching signal along the two-dimensional (surface) target. We will call the correlation function Rβ(∆, ∆β0) as the azimuth-normalized correlation function of the target return signal fluctuations and the correlation function Rγ(∆, ∆γ0, t, τ) as the aspect-angle-normalized correlation function of the target return signal fluctuations.

Copyright 2005 by CRC Press

1598_book.fm Page 68 Monday, October 11, 2004 1:57 PM

68 2.5.3

Signal and Image Processing in Navigational Systems The Target Return Signal Power with the Pulsed Searching Signal

The target return signal power as a function of time is given by Equation (2.136). Taking into consideration that

∫ g ( ϕ , ψ ) dϕ = ∆

(2)

2 h

*

h

(ϕ * ) = ∆(h2)

(2.137)

*

and using the generalized theorem about the mean, we can write

p= =

( ) (

)

(2) PSG02 λ 2 ∆ h* ∆ p gv2 ψ * S° β 0 , γ * sin γ * 64 π h

3 2

(

) (

PSG02 λ 2 kah ∆ h* kp cτ p gv2 γ * − γ 0 S° β 0 , γ *

)

,

(2.138)

128 π 3 ρ*3 cos γ *

where ρ* =

∆p =

∫

h h = 0.5 ct ; = sin γ * sin( γ 0 + ψ * )

Π2 t − 

kp cτ p sin 2 γ * kp cτ p tgγ * ( ) d ψ = = c   2 h cos γ * 2ρ*

2ρ ψ

(2.139)

(2.140)

is the interval of variations of the angle ψ or γ within the effective duration of the pulsed searching signal squared envelope τ (p2 ) = k p τ p ;

(2.141)

τp is the effective duration of the pulsed searching signal and kp is the coefficient of the pulsed searching signal shape. Under the condition γ* ≈ 90° the equality in Equation (2.141) is not true. In the determination of ∆p, we used the following approximation t−

2ρ 2(ρ* − ρ) 2ρ( ψ − ψ * ) = ≈ ⋅ ctg γ * . c c c

(2.142)

Comparing Equation (2.107) and Equation (2.138), we can define the effective scattering area of the two-dimensional (surface) target as follows:

(

) (

)

Stsurface = gv2 γ * − γ 0 ⋅ S° β 0 , γ * ⋅ Sarea , Copyright 2005 by CRC Press

(2.143)

1598_book.fm Page 69 Monday, October 11, 2004 1:57 PM

Correlation Function of Target Return Signal Fluctuations

69

where Sarea =

k ah ∆ h* ρ* k p cτ p

(2.144)

2 cos γ *

is the geometrical area of the two-dimensional (surface) target resolution element. Thus, under scanning of the two-dimensional (surface) target by shortduration pulsed searching signals, i.e., when the condition ∆p << ∆(v2 ) is satisfied, the target return signal power at the instant of time t, for the distance ρ (the radar range) ρ* = 0.5 ct,

(2.145)

is proportional to the amplification factor of the vertical-coverage radar antenna directional diagram under the condition ψ* = γ* – γ0, and with the fixed angles γ* and γ0 is inversely proportional to ρ3* . This can be explained by the fact that the area of a two-dimensional (surface) target resolution element is proportional to ρ (the radar range). Dependences of the received target return signal power on the radar range are shown in Figure 2.10 when the directional diagram is considered Gaussian and S° = const. The obtained results are true for a wide range of angles, including angles that are very close to 90°. p 1

1.00

0.75

−h = 5 km −h = 10 km

2 0.50 1 3

0.25

2 4 0

10

20

30

40

50

3 60

ρ (km)

4 70

80

90

100 110 120 130 140

FIGURE 2.10 The target return signal power as a function of radar range: (1) γ0 = 12°; (2) γ0 = 8°; (3) γ0 = 4°; (4) γ0 = 0°.

Copyright 2005 by CRC Press

1598_book.fm Page 70 Monday, October 11, 2004 1:57 PM

70

Signal and Image Processing in Navigational Systems

During the process of propagation of the pulsed searching signal, the power p(t) of the target return signal is a very complex function of time t = c * . This is one source of the nonstationary state of the target return signal at the receiver or detector input in navigational systems. It is not difficult to define exactly the target return signal power on the basis of Equation (2.138). For this purpose, it is necessary to introduce the (γ*) = 2ρ

. arcsin 2h ct But in many cases, the simple approximation is sufficient; when the vertical-coverage directional diagram is not large, i.e., when the condition sin ∆v ≈ ∆v is satisfied, and when the angle γ0 is not small, i.e., the condition ∆v ctg γ0 << 1 is true, the linear function ψ* = γ * − γ 0 =

Td − t Td ctg γ 0

(2.146)

or t = Td(1 – ψ* ctg γ0)

(2.147)

is true, where Td =

2ρ0 2h = c c sin γ 0

(2.148)

is the delay of the target return signal from the surface of the two-dimensional target on the directional diagram axis.63,64 Substituting Equation (2.147) in Equation (2.138) and taking into consideration Equation (2.145) and the equality cos γ * = 1 −

h2 , ρ*2

(2.149)

we can define the target return signal power, which depends on the specific functions gh(ψ*) and S°(β0, γ*). For example, in the case of the Gaussian vertical-coverage directional diagram and the exponential dependence of the function S°(β0, γ*) discussed in Feldman and Mandurovsky27

(

)

(

)

(

)

k β ,γ ψ S° β 0 , γ * = S° β 0 , ψ + γ 0 = S° β 0 , γ 0 ⋅ e 1( 0 0 ) ,

(2.150)

where k1 is a coefficient defined by the landscape and the angles β0 and γ0, the target return signal power has the following form:

Copyright 2005 by CRC Press

1598_book.fm Page 71 Monday, October 11, 2004 1:57 PM

Correlation Function of Target Return Signal Fluctuations

() ( )

p t = p Td ⋅

3 d 3

−π ⋅

(t − Td )2 − k ∆

T e ⋅ t 1−

1 h

Tτ2

Td2 t2

⋅

71

t − Td 2 Tτ

⋅ sin 2 γ 0

,

(2.151)

where Tτ =

∆h 2

⋅ Td ctg γ 0

(2.152)

is the duration of the target return signal, which is defined as the difference of signal delays from the points of the vertical-coverage directional diagram corresponding to the angles γ0 ± 0.5∆(2) h , and

( )

p Td =

(

PSG02 λ 2 ∆ h cτ p kah kpS° β 0 , γ 0 128 π ρ cos γ 0 3 3 0

).

(2.153)

The target return signal power given by Equation (2.151) has the shape of the vertical-coverage directional diagram, but is deformed under the influence of the functions S°(ψ) and ρ–3, and its maximum is shifted. When T Equation (2.146) and Equation (2.147) are true, the value of td is not so high, and we can write p(t) ≈ p1 (Td ) ⋅ e

− π⋅

( t −Td′ )2 Tτ2

,

(2.154)

;

(2.155)

where p1 (Td ) = p(Td ) ⋅ e Td′ = Td  1 − 

Q∆(h ) 4π

(2) h 8π

Q2∆

⋅ ctg γ 0  ; 

(2.156)

Q = k1 + 3 ctg γ0 + tg γ0.

(2.157)

2

The relative shift in maximum of the target return signal power is determined as follows: δp =

Copyright 2005 by CRC Press

Td − Td′ ∆(v2 ) = ⋅ ( k1 + 3 ctg γ 0 + tg γ 0 ) ⋅ ctg γ 0 . Td 4π

(2.158)

1598_book.fm Page 72 Monday, October 11, 2004 1:57 PM

72

Signal and Image Processing in Navigational Systems TABLE 2.1 Various Kinds of Surfaces of the Two-Dimensional Target Surface

k1

Note

Forest Plow and dry snow Dry sand Ice Sea under conditions of wind velocity: 48 km/h 18 … 28 km/h 0 … 11 km/h

0.62 3.3 5 6.5

— λ = 3 cm γ0 = 50 … 80˚

6.7 10.7 16

— — —

The shift δp of the target return signal power can be defined on the basis of Equation (2.138) and Equation (2.139) under the condition p′(γ*) = 0, when the maximum of the function p(γ*) is known. The first term in Equation (2.158) is defined by the function S°(γ). The second and third terms in Equation (2.158) are defined by the radar range. When the surface of the two-dimensional target is smooth (see Table 2.1) and the condition k1 > 3 ctg γ0 + tg γ0 is satisfied, the main reason for the shift in the maximum target return signal power is the function S°(γ). When the surface of the two-dimensional target is very rough, the main reason for the shift in maximum power is a function of the radar range. For example, at γ0 = 45°, ∆h = 120°, and k1 = 16, which corresponds to the quiescent surface of the sea, we obtain δp = (5 + 1 + 0.3)% and at k1 = 0.6, which corresponds to the forest, we obtain δp = (0.2 + 1 + 0.3)%. When the directional diagram is Gaussian and the function S°(γ) is given by Equation (2.150), the target return signal power can be defined on the basis of Equation (2.124) without assuming that the pulsed searching signal has a short duration in comparison with Tr . Consider the condition τp >> Tr . In this case, the target return signal has a shape that is very close to a square waveform. The duration of the target return signal is equal to τp. The durations of the leading and trailing edges of the target return signal are very close to Tr . The maximum target return signal power, when the condition sinγ0 ≈ sin(ψ + γ0) is satisfied, is determined as follows:

( )

Pmax = p Td′′ =

(

)

PSG02 λ 2 ∆ h ∆ vS° β 0 , γ 0 sin γ 0 128 π 3 h2

()

⋅e

2 k12 ∆ v 8π

,

(2.159)

where Td′′= Td  1 −

Copyright 2005 by CRC Press

k1∆(v2 ) 4π

⋅ ctg γ 0  .

(2.160)

1598_book.fm Page 73 Monday, October 11, 2004 1:57 PM

Correlation Function of Target Return Signal Fluctuations

73

Under the conditions h = variable and γ0 = const, the maximum target return signal power is inversely proportional to h2 or ρ2. When the conditions h = const and γ0 = variable are satisfied, the maximum target return signal power is inversely proportional to ρ. The error in the definition of the target return signal power caused by the functional dependence in Equation (2.151) is usually not high for high values of k1. For example, at k1 = 16 and ∆v = 6˚, this error is approximately equal to 10%. It is not difficult to define the target return signal power for arbitrary values τp and Tr if the conditions in Equation (2.142) and Equation (2.150) are satisfied and the vertical-coverage directional diagram and the pulsed searching signal are Gaussian. Neglecting Equation (2.150) for simplicity, we can write

p=

( ) (

)

PSG02 λ 2 ∆ h gv2 ψ * S° β 0 , γ 0 sin γ 0 128 π h

3 2

⋅I ,

(2.161)

where

I=

∆v ∆p ⋅e 2 2 ∆( ) + ∆( ) v

−2 π ⋅

ψ* 2 2 ∆v + ∆ p

() ()

.

(2.162)

p

Then, in the case of the short-duration pulsed searching signal, i.e., when the condition ∆p << ∆v is satisfied, we can write 2

I = ∆p ⋅ e

ψ −2 π⋅ *

(2) ∆v

(2.163)

and Equation (2.161) is transformed to Equation (2.138). In the case of the long-duration pulsed searching signal, i.e., when the condition ∆p >> ∆v is satisfied, we can assume that γ* = γ0 or ψ* = 0 and I = ∆v . In this case, Equation (2.161) is transformed to Equation (2.159), taking into consideration the condition k1 = 0. Thus, in the case of the short-duration pulsed searching signal, the target return signal power is inversely proportional to ρ3* where ρ* is the distance between the radar and the center of a two-dimensional (surface) target resolution element [see Equation (2.138)]. In the case of the long-duration pulsed searching signal, when the signal completely covers the scanned surface of the two-dimensional target, the target return signal power is inversely proportional to ρ0 where ρ0 is the distance between the radar and center of the scanned surface, since [see Equation (2.159)]

Copyright 2005 by CRC Press

1598_book.fm Page 74 Monday, October 11, 2004 1:57 PM

74

Signal and Image Processing in Navigational Systems

sin γ 0 =

2.5.4

h . ρ0

(2.164)

The Correlation Function and Power of the Target Return Signal with the Simple Harmonic Searching Signal

Assuming that the condition given by Equation (2.109) is satisfied in Equation (2.122) and omitting the summation sign, we can write

( )

(

)

R∆enρ, ∆ϕ , ∆ψ t , τ = R en ∆, ∆β 0 , ∆γ 0 = p0

(

∫∫ g (ϕ − 0.5∆∆ϕ, ψ − 0.5∆ψ )

× g ϕ + 0.5 ∆ϕ , ψ + 0.5 ∆ψ

(

)

) (

)

× S° β 0 , ψ + γ 0 sin ψ + γ 0 ⋅ e

−2 jω 0 ⋅

(

∆ρ ϕ , ψ c

)

dϕ dψ (2.165)

where the differentials are given by Equation (2.125)–Equation (2.127), respectively, and it is necessary to replace γ* with γ0 and ρ* with ρ0, respectively, using Equation (2.164). Equation (2.165), as well as Equation (2.133), can be represented as the product of the azimuth and aspect angle normalized correlation functions of target return signal fluctuations if the condition in Equation (2.97) is satisfied. If the condition ∆ρ = ∆ϕ = ∆ψ = 0 is satisfied in Equation (2.165), we can define the target return signal power in the following form:

()

p t = p0

∫∫ g (ϕ, ψ ) ⋅ S° (β , ψ + γ ) sin ( ψ + γ ) dϕ dψ . 2

0

0

0

(2.166)

If the radar antenna directional diagram width is not so large and the variables in the function g(ϕ, ψ) are separated, then when the conditions S°(β0, ψ + γ0) ≈ S°(β0, γ*) and sin(ψ + γ) ≈ sin γ* are satisfied, we can write p=

(

)

(2) 2 PSG02 λ 2 ∆ h ∆ (v )S° β 0 , γ 0 sin γ 0 64 π h

3 2

.

(2.167)

Comparing Equation (2.167) and Equation (2.107), we can define the effective scattering area of the two-dimensional (surface) target in the following form:

(

)

Stsurface = S° β 0 , γ 0 ⋅ Ssurface ,

(2.168)

where Ssurface =

Copyright 2005 by CRC Press

(2) 2 ∆ h ∆ (v )ρ20 sin γ 0

(2.169)

1598_book.fm Page 75 Monday, October 11, 2004 1:57 PM

Correlation Function of Target Return Signal Fluctuations

75

is the surface covered by the directional diagram. Equation (2.167) is equivalent to Equation (2.159). Equation (2.167) is true within the wide range of variation of the angle γ0 except for very small values of the angle γ0, when the top edge of the directional diagram breaks away from the surface of the two-dimensional target. Equation (2.167) can also be used for high values of the directional diagram width, but the directional diagram should be uniformly contrasting.

2.6

The Correlation Function under Vertical Scanning of the Two-Dimensional (Surface) Target

Under scanning of the two-dimensional (surface) target at angles that are very close to 90° it is convenient to introduce a new coordinate system, in which the position of some scatterer D on the surface is given by the angles α and θ (see Figure 2.11). Assuming that the deviation θ0 of the radar antenna directional diagram axis from the vertical line and the directional diagram width are infinitesimal, we can use the approximate equality: sinθ ≈ θ.65,66 The origin of the coordinate system OXYZ is matched with a view of the phase center of the radar antenna in navigational systems so that z = h, and the axes OX and OY are directed in parallel to orthogonal straight lines that are formed at the intersection of the surface by the main planes ϕ and ψ of the two-dimensional directional diagram. Under these conditions, we can write ϕ = θ cos α – ϕ0 ρ=

and ψ = θ sin α – ϕ0;

(

)

h ≈ h 1 + 0.5 θ 2 , cos θ

(2.170) (2.171)

where ϕ0 = θ0 cos α0; ψ0 = θ0 sin α0

and dϕdψ = sin θ dθ dα.

(2.172)

Let us assume that the radar moves uniformly and linearly with velocity V the direction of which is given by the azimuth angle β0 in the horizontal plane with respect to the axis OX and the trajectory angle ε0 in the vertical plane.67 Then ∆ρ(α, θ) = –Vr(α, θ) τ = –V · [cos ε0 cos(α – β0) sin θ + sin ε0 cos θ]τ. (2.173) The product of the pulsed integrands in Equation (2.122) for the given instant of time t defines the interval of integration with respect to the variable Copyright 2005 by CRC Press

1598_book.fm Page 76 Monday, October 11, 2004 1:57 PM

76

Signal and Image Processing in Navigational Systems

Y

ϕ0

ϕ

θ

ψ

θ0

h

y

ψ0

D S

α

α0

X x

FIGURE 2.11 The coordinate system using the variables α and θ.

θ. With square waveform pulsed searching signals of duration τp as follows from the condition of overlapping

(

)

− 0.5 τ p − τ ≤ t −

2ρ ≤ 0.5 τ p − τ , c

(

)

(2.174)

the limits of integration can be determined as follows

(

)

θ1,2 = θ *2 ∓ τ p − µτ + nTp ⋅

c = 2h

(

2(t − Td ) ∓ τ p − µτ + nTp Td

),

(2.175)

where θ *2 ≈

2(ρ* − h) h

≈

2(t − Td ) Td

(2.176)

and ρ* =

Copyright 2005 by CRC Press

(

h ≈ h 1 + 0.5 θ *2 cos θ *

)

(2.177)

1598_book.fm Page 77 Monday, October 11, 2004 1:57 PM

Correlation Function of Target Return Signal Fluctuations

77

∆a 2a

θ∗ ∆a2 h

θ2 θ1

ρ∗ c τp

4a 2

2

FIGURE 2.12 Space relationships under vertical scanning of surface.

are the angle and distance defining the position of the considered resolution element at the instant of time t = the target return signal;

2ρ* c

µ = 1+

(see Figure 2.12); Td =

2h c

2V ⋅ cos θ * . c

is the delay of

(2.178)

Clearly, the following condition t – Td ≥ 0.5(τp – µτ + nTp)

(2.179)

should be satisfied. For the instants of time that are within the interval [Td – 0.5(τp – µτ + nTp), Td + 0.5(τp – µτ + nTp)],

(2.180)

the condition θ1 = 0 is true. If the condition t < Td – 0.5(τp – µτ + nTp) is satisfied, we can write θ1 = θ2 = 0. Finally, on the basis of Equation (2.122) we can write Copyright 2005 by CRC Press

(2.181)

1598_book.fm Page 78 Monday, October 11, 2004 1:57 PM

78

R

Signal and Image Processing in Navigational Systems

en

∞

( )

t , τ = p0

()

2 π θ2 n

∑ ∫ ∫ g (θ cos α − ϕ , θ sin α − ψ ) ⋅ S° (θ) ⋅ e () 2

0

0

( ) sin θ dθ dα ,

jΩ α ,θ τ

n= 0 0 θ n 1

(2.182) where p0 =

PSG02 λ 2 64 π 3 h2

(2.183)

4 πVr (α , θ) . λ

(2.184)

and Ω(α , θ) =

Assuming τ = 0 and n = 0 in Equation (2.182), we can define the target return signal power. It is supposed that the two-dimensional target has a rough surface and that the coherent component of the target return signal is absent.68 So,

()

p t = p0

2 π θ2

∫ ∫ g (θ cos α − ϕ , θ sin α − ψ ) ⋅ S° (θ) sin θ dθ dα , 2

0

0

(2.185)

0 θ1

where θ1,2 = θ *2 ∓

cτ p 2h

=

2(t − Td ) ∓ τ p Td

,

t > Td + 0.5τ p .

(2.186)

Let us assume that the directional diagram has symmetric axes and obeys the Gaussian law. Then, the effective scattering area as a function of the angle θ is Gaussian too,27 and

()

2

S° θ = SNo ⋅ e − k2θ ,

(2.187)

where SNo is the effective scattering area under vertical scanning. The function in Equation (2.188), when the angle θ is not so high in value, is equivalent to the function determined by:69

()

2

S° θ = SNo ⋅ e − k2tg θ .

Copyright 2005 by CRC Press

(2.188)

1598_book.fm Page 79 Monday, October 11, 2004 1:57 PM

Correlation Function of Target Return Signal Fluctuations

79

The parameter k2 characterizes the roughness of the surface of the twodimensional target and increases as the roughness of the surface decreases. Then we can write27 2 π θ2

()

p t = p0SNo

∫ ∫e

−2 π⋅

a 2 θ2 + θ20 − 2 θθ0 cos α

()

2 ∆a

θ dθ da ,

(2.189)

0 θ1

where

a2 = 1 +

2 k2 ∆ (a ) 2π

(2.190)

and ∆a is the directional diagram width in navigational systems. Integrating with respect to the variable a and introducing a new variable ∆a

θ=

2 πα

⋅x

x=

or

2 πa ⋅θ , ∆a

(2.191)

we can determine the target return signal power in the following form:

()

p t =

2 PSG02 λ 2 ∆ (a )SNo ⋅F t , 128 π 3 a2 h2

()

(2.192)

) I (bx)dx ; 0

(2.193)

where F (t ) =

x2 ( t )

∫ x⋅e

(

−0.5 x 2 + b 2

x1 ( t )

()

x1,2 t = 2 π a ⋅

t − Td ∓ 0.5τ p θ1,2 = 2π ⋅ , ∆a Tpcon b=2 π⋅

t > Td + 0.5τ p ;

θ0 . a∆ a

(2.194)

(2.195)

One important point to remember is that if Td – 0.5τp < t < Td + 0.5τp,

Copyright 2005 by CRC Press

then x1 = 0

(2.196)

1598_book.fm Page 80 Monday, October 11, 2004 1:57 PM

80

Signal and Image Processing in Navigational Systems

and if t < Td – 0.5τp,

then x1 = x2 = 0

and F(t) = 0.

(2.197)

Taking into consideration Equation (2.196) and Equation (2.197), we can write Tpcon =

∆(a2 ) h∆(a2 ) ⋅ T = , d 4a2 2a2 c

(2.198)

where Tpcon is the conditional duration of the target return signal, equal to the time during which the pulse leading edge is propagated along a surface from the first point of tangency t = Td – 0.5τp to the circle observed under ∆ ∆ the angle aa (see Figure 2.13 and Figure 2.14); aa is the width of the equivalent directional diagram that is more narrow because of the function S° (θ) given by Equation (2.187); a is the coefficient of narrowing given by Equation (2.190). Changes in the conditional duration Tpcon of the target return signal given by Equation (2.198) can occur in the case of two-dimensional targets with smooth surfaces. For example, at ∆a = 12° and k2 = 200 — the case of weak sea waves — Tpcon is decreased 2.5 times.27,69 F (t ) 1.0

1

0.8

2

0.6 3

4

0.4

5

6

0.2

t −Td Tpcon 0

1

2

FIGURE 2.13 The function F(t) vs. the ratio θ 1.5; (6) ∆ = 2. 0

a

Copyright 2005 by CRC Press

3 θ0 ∆a

4

at

τp con

Tp

5

= 2: (1)

6 θ0 ∆a

7

= 0; (2)

8 θ0 ∆a

= 0.5; (3)

9 θ0 ∆a

10

11

= 0.75; (4)

θ0 ∆a

= 1; (5)

θ0 ∆a

=

1598_book.fm Page 81 Monday, October 11, 2004 1:57 PM

Correlation Function of Target Return Signal Fluctuations

81

F (t ) 1

1.0

2 0.8

0.6

3

4

0.4

0.2 t −Td Tpcon 0

1

2

FIGURE 2.14 The function F(t) vs. the ratio

τp

3

at θ0 = 0: (1)

con

Tp

4 τp con

Tp

= 2; (2)

5 τp con

Tp

= 1; (3)

6 τp con

Tp

= 0.5; (4)

τp con

Tp

= 0.33.

The integral in Equation (2.193) can be determined using the incomplete Toronto function:27

()

(

F t = T x2 1, 0, 2

b 2

) − T (1, 0, ) . b 2

x1 2

(2.199)

In the particular case where θ0 = 0,27 we can write 2

1− e

x − 2

2

= 1− e

− π⋅

t − Td + 0.5 τ p Tpcom

(2.200)

if the condition in Equation (2.196) is satisfied and

()

F t =e

−0.5 x12

−e

−0.5 x22

=e

−π ⋅

t − Td − 0.5 τ p Tpcom

−e

−π ⋅

t − Td + 0.5 τ p Tpcom

(2.201)

if the condition t > Td + 0.5τp is true. The function F(t) the shape of the target return signal from the two-dimensional (surface) target during the time interval t – Td. The function F(t) is shown in Figure 2.13 and Figure 2.14 for some values of

θ0 ∆a

and

τp Tpcon

. In particular, for θ0 = 0, the maximum of the

function F(t) is determined by

Copyright 2005 by CRC Press

1598_book.fm Page 82 Monday, October 11, 2004 1:57 PM

82

Signal and Image Processing in Navigational Systems τp

Fmax (t) = 1 − e

− π⋅ com T p

(2.202)

and exists if the following condition t = Td + 0.5τp is satisfied. With the continuous searching signal, when the conditions x1 = 0, x2 → ∞, and F(t) → 1 are satisfied, the target return signal power has the following form:

()

p t =

2.7

2 PSG02 λ 2 ∆ (a )SNo . 128 π 3 a2 h2

(2.203)

Conclusions

The discussion in this chapter allows us to draw the following conclusions. The physical sources of target return signal fluctuations are defined. Sources of the fluctuations can be considered as changes in the amplitude, phase, or frequency of elementary signals that lead to variations in the amplitude, phase, or frequency of the resulting target return signal. The amplitude, phase, or frequency of elementary signals can be changed as a consequence of: displacements and rotation of scatterers under the stimulus of the wind; high angle dimensions of the two-dimensional (surface) or three-dimensional (space) target; Doppler frequencies and the secondary Doppler effect; antenna scanning or rotation of the scanning polarization plane of radar antenna; and the nonstationary state of the searching signal frequency. The following are forms of target return signal fluctuations: fluctuations in the radar range; interperiod fluctuations; intraperiod fluctuations; slow fluctuations; rapid fluctuations; time fluctuations; and space fluctuations. The target return signal can be characterized by the normalized correlation function of time and space fluctuations. The two-dimensional spectral power density can define the nonstationary target return signal. The one-dimensional spectral power density defines the stationary target return signal. In the case of the narrow-band searching signal, the correlation function of the target return signal fluctuations is the correlation function of the slow fluctuations with the continuous searching signal. In other words, this is the correlation function of the fluctuations caused by the moving radar, antenna scanning, and displacements of scatterers — i.e., the space fluctuations. With the pulsed searching signal, the correlation function of target return signal fluctuations is defined both by the fluctuations in the radar range — the intraperiod fluctuations — and by the interperiod fluctuations. The generalized normalized correlation function of target return signal fluctuations under scanning of the three-dimensional (space) target is

Copyright 2005 by CRC Press

1598_book.fm Page 83 Monday, October 11, 2004 1:57 PM

Correlation Function of Target Return Signal Fluctuations

83

defined by the product of two normalized correlation functions. The first normalized correlation function defines the slow fluctuations caused by the moving radar, displacements of scatterers, and antenna scanning and the rapid fluctuations caused by propagation of the searching signal in space. The second normalized correlation function defines the slow target return signal fluctuations caused by rotation of scatterers and the radar antenna polarization plane. In the case of the pulsed searching signal, the normalized correlation function of the fluctuations that are caused by the moving radar and antenna scanning, too, is defined by the product of two normalized correlation functions. The first normalized correlation function defines the slow fluctuations caused by the different Doppler shifts in the frequency of elementary signals within the pulsed searching signal resolution area with moving radar, and amplitude changes in elementary signals caused by antenna scanning, and variations of the aspect angle during radar motion — the interperiod fluctuations. The second normalized correlation function is a periodic function with respect to τ and defines the rapid fluctuations caused by propagation of the pulsed searching signal in space with respect to simultaneous radar motion — i.e., the intraperiod fluctuations. Under scanning of the three-dimensional (space) target as well as a point target by the pulsed searching signal, the target return signal power is inversely proportional to ρ2, where ρ is the radar range, but not inversely proportional to ρ4. This can be explained by the fact that with an increase in the radar range, the area of the pulsed searching signal and the effective scattering area increase proportionally to ρ2. Under angle scanning of the two-dimensional (surface) target by the pulsed searching signal, the normalized correlation function of the fluctuations is defined by the product of the azimuth-normalized correlation function Rβ(∆, ∆β0) and the aspect-anglenormalized correlation function Rγ(∆, ∆γ0, t, τ). The azimuth-normalized correlation function defines the slow fluctuations caused by the moving radar with varying phase changes in elementary signals in the azimuth plane and by rotation of the radar antenna axis. The aspect-angle-normalized (or space–time) correlation function defines the slow fluctuations caused by the moving radar with varying phase changes in elementary signals in the aspect angle plane, and by rotation of the radar antenna axis, and defines the rapid fluctuations caused by propagation of the pulsed searching signal along the two-dimensional (surface) target. Under vertical scanning of the two-dimensional (surface) target by the pulsed searching signal, the normalized correlation function of the fluctuations and power of the target return signal are defined on the basis of the Toronto function. The main assertions that are true under angle scanning of the two-dimensional (surface) target are also true under vertical scanning.

Copyright 2005 by CRC Press

1598_book.fm Page 84 Monday, October 11, 2004 1:57 PM

84

Signal and Image Processing in Navigational Systems

References 1. Feldman, Yu, Gidaspov, Yu, and Gomzin, V., Moving Target Tracking, Soviet Radio, Moscow, 1978 (in Russian). 2. Bacut, P. et al., Problems of Statistical Radar, Soviet Radio, Moscow, 1963 (in Russian). 3. Dulevich, V. et al., Theoretical Foundations of Radar, Soviet Radio, Moscow, 1978 (in Russian). 4. Ferrara, E. and Parks, T., Direction finding with an array of antennas having diverse polarizations, IEEE Trans., Vol. AP-31, No. 3, 1983, pp. 231–236. 5. Roy, R. and Kailath, T., ESPRIT — Estimation of signal parameters via rotational invariance techniques, IEEE Trans., Vol. ASSP-37, No. 7, 1989, pp. 984–995. 6. Wong, K. and Zoltowski, M., High accuracy 2D angle estimation with extended aperture vector sensor arrays, in Proceedings of the ICASSP, Vol. 5, May 1996, pp. 2789–2792. 7. Kolchinsky, V., Mandurovsky, I., and Konstantinovsky, M., Doppler Devices and Navigational Systems, Soviet Radio, Moscow, 1975 (in Russian). 8. Li, J., Direction and polarization estimation using arrays with small loops and short dipoles, IEEE Trans., Vol. AP-41, No. 3, 1993, pp. 379–387. 9. Winitzky, A., Basis of Radar under Continuous Generation of Radio Waves, Soviet Radio, Moscow, 1961 (in Russian). 10. Farina, A., Gini, F., Greco, M., and Lee, P., Improvement factor for real-sea clutter Doppler frequency spectra, in Proc. Inst. Elect. Eng. F, Vol. 123, October, 1996, pp. 341–344. 11. Cirban, H. and Tsatsanis, M., Maximum likelihood blind channel estimation in the presence of Doppler shifts, IEEE Trans., Vol. SP-47, No. 5, 1999, pp. 1559–1569. 12. Rytov, S., Introduction to Statistical Radio Physics. Part I: Stochastic Processes, Nauka, Moscow, 1976 (in Russian). 13. Blackman, S., Multiple-Target Tracking with Radar Applications, Artech House, Norwood, MA, 1986. 14. Gardner, W., Introduction to Random Processes with Application to Signals and Systems, 2nd ed., McGraw-Hill, New York, 1989. 15. Richaczek, A., Principles of High-Resolution Radar, Peninsula, San Francisco, CA, 1985. 16. Abarband, H., Analysis of Observed Chaotic Data, Springer-Verlag, New York, 1996. 17. Zukovsky, A., Onoprienko, E., and Chizov, V., Theoretical Foundations of Radio Altimetry, Soviet Radio, Moscow, 1979 (in Russian). 18. Verdu, S., Multiuser Detection, Cambridge University Press, Cambridge, U.K., 1988. 19. Rappaport, T., Wireless Communications: Principles and Practice, Prentice Hall, Englewood Cliffs, NJ, 1996. 20. Papoulis, L., Signal Analysis, McGraw-Hill, New York, 1977. 21. Tikhonov, V., Statistical Radio Engineering, Radio and Svyaz, Moscow, 1982 (in Russian). 22. Yaglom, A., Correlation Theory of Stationary Stochastic Functions, Hidrometeoizdat, Saint Petersburg, 1981 (in Russian).

Copyright 2005 by CRC Press

1598_book.fm Page 85 Monday, October 11, 2004 1:57 PM

Correlation Function of Target Return Signal Fluctuations

85

23. Kay, S., Fundamentals of Statistical Signal Processing: Detection Theory, Vol. 2, Prentice Hall, Englewood Cliffs, NJ, 1998. 24. Macchi, O., Adaptive Processing, John Wiley & Sons, New York, 1995. 25. Scharf, L., Statistical Signal Processing: Detection, Estimation and Time Series Analysis, Addison-Wesley, Reading, MA, 1991. 26. Stoica, P. and Moses, R., Introduction to Spectral Analysis, Prentice Hall, Upper Saddle River, NJ, 1997. 27. Feldman, Yu and Mandurovsky, I., Theory of Fluctuations of Radar Signals, Radio and Svyaz, Moscow, 1988 (in Russian). 28. Widrow, B. and Stearns, S., Adaptive Signal Processing, Prentice Hall, Englewood Cliffs, NJ, 1985. 29. Poor, H., An Introduction to Signal Detection and Estimation, 2nd ed., SpringerVerlag, New York, 1994. 30. Rytov, S., Kravtzov, Yu, and Tatarsky, V., Introduction to Statistical Radio Physics. Part II: Stochastic Fields, Nauka, Moscow, 1978 (in Russian). 31. Cohen, L., Time-Frequency Analysis, Prentice Hall, Englewood Cliffs, NJ, 1995. 32. Shuster, H., Deterministic Chaos, VCH, New York, 1989. 33. Hannan, E. and Deistler, M., The Statistical Theory of Linear Systems, John Wiley & Sons, New York, 1988. 34. Jain, A., Fundamentals of Digital Image Processing, Prentice Hall, Englewood Cliffs, NJ, 1989. 35. Haykin, S., Adaptive Filter Theory, 3rd ed., Prentice Hall, Upper Saddle River, NJ, 1996. 36. Kay, S., Modern Spectral Estimation: Theory and Application, Prentice Hall, Englewood Cliffs, NJ, 1988. 37. Paholkov, G., Cashinov, V., and Ponomarenko, B., Variational Technique for Synthesis of Signals and Filters, Radio and Svyaz, Moscow, 1981 (in Russian). 38. Kassam, S., Signal Detection in Non-Gaussian Noise, Springer-Verlag, New York, 1988. 39. Gerlach, K. and Steiner, M., Adaptive detection of range distributed targets, IEEE Trans., Vol. SP-47, No. 7, 1999, pp. 1844–1851. 40. Oppenheim, A. and Willsky, A., Signals and Systems, Prentice Hall, Englewood Cliffs, NJ, 1983. 41. Proakis, J., Digital Communications, 3rd ed., McGraw-Hill, New York, 1995. 42. Van Trees, H., Detection, Estimation and Modulation Theory. Part III: Radar-Sonar Signal Processing and Gaussian Signals in Noise, John Wiley & Sons, New York, 1972. 43. Porat, B., Digital Processing of Random Signals: Theory and Methods, Prentice-Hall, Englewood Cliffs, NJ, 1994. 44. Davies, K., Ionospheric Radio, Peter Peregrinns, London, 1990. 45. Friedlander, B. and Francos, A., Estimation of amplitude and phase parameters of multicomponent signals, IEEE Trans., Vol. SP-43, No. 4, pp. 917–927. 46. Jeruchim, M., Balaban, P., and Shanmugan, K., Simulation of Communication Systems, Plenum, New York, 1992. 47. McNamara, L., The Ionosphere: Communications, Surveillance, and Direction Finding, Krieger, Malabar, FL, 1991. 48. Gingras, D., Gerstoft, P., and Gerr, N., Electromagnetic matched-field processing: basic concepts and tropospheric simulations, IEEE Trans., Vol. AP-45, No. 10, 1997, pp. 1536–1545.

Copyright 2005 by CRC Press

1598_book.fm Page 86 Monday, October 11, 2004 1:57 PM

86

Signal and Image Processing in Navigational Systems

49. Micka, O. and Weiss, A., Estimating frequencies of exponentials in noise using joint diagonalization, IEEE Trans., Vol. SP-47, No. 2, 1999, pp. 341–348. 50. Francos, A. and Porat, M., Analysis and synthesis of multicomponent signals using positive time–frequency distributions, IEEE Trans., Vol. SP-47, No. 2, 1999, pp. 493–504. 51. Conte, E., Di Bisceglie, M., Longo, M., and Lops, M., Canonical detection in spherically invariant noise, IEEE Trans., Vol. COM-43, No. 2, 1995, pp. 347–353. 52. Therrien, C., Discrete Random Signals and Statistical Signal Processing, Prentice Hall, Englewood Cliffs, NJ, 1992. 53. Kapoor, S., Marchok, D., and Huang, Y.-F., Adaptive interference suppression in multiuser wireless OFDM systems using antenna arrays, IEEE Trans., Vol. SP-47, No. 12, 1999, pp. 3381–3391. 54. Trump, T. and Ottersten, B., Estimation of nominal direction of arrival and angular spread using an array of sensors, Signal Process., Vol. 50, No. 1–2, 1996, pp. 57–69. 55. Anderson, C., Green, S., and Kingsley, S., HF skywave radar: estimating aircraft heights using super-resolution in range, in Proc. Inst. Elect. Eng. Radar Sonar Navigat., Vol. 143, August 1996, pp. 281–285. 56. Nehorai, A., Ho, K.-C., and Tan, B., Minimum-noise-variance beam former with an electromagnetic vector sensor, IEEE Trans., Vol. SP-47, No. 3, 1999, pp. 601–618. 57. Anderson, T., An Introduction to Multivariate Statistical Analysis, 2nd ed., John Wiley & Sons, New York, 1984. 58. Leung, H. and Lo, T., Chaotic radar signal processing over the sea, IEEE J. Oceanic Eng., Vol. 18, 1993, pp. 287–295. 59. Godard, D., Self-recovering equalization and carrier tracking in two-dimensional data communication systems, IEEE Trans., Vol. COM-32, No. 6, 1975, pp. 679–682. 60. Agafe, C. and Iltis, R., Statistics of the RSS estimation algorithm for gaussian measurement noise, IEEE Trans., Vol. SP-47, No. 1, 1999, pp. 22–32. 61. Wong, K. and Zoltowski, M., Univector-sensor ESPRIT for multi-source azimuth-elevation angle-estimation, in Antennas and Propagation Society International Symposium, Vol. 2, AP-S. Digest, 1996, pp. 1368–1371. 62. Papazoglou, M. and Krolik, J., Matched-field estimation of aircraft altitude from multiple over-the-horizon radar revisits, IEEE Trans., Vol. SP-47, No. 4, 1999, pp. 966–975. 63. Barbarossa, S. and Scaglione, A., Adaptive time-varying cancellation of wideband interferences in spread-spectrum communications based on time-frequency distributions, IEEE Trans., Vol. SP-47, No. 4, pp. 957–965. 64. Frenkel, L. and Feder, M., Recursive expectation maximization (EM) algorithms for time-varying parameters with applications to multiple target tracking, IEEE Trans., Vol. SP-47, No. 2, 1999, pp. 306–320. 65. Tufts, D. and Kumaresan, R., Estimation of frequencies of multiple sinusoids: making linear prediction perform like maximum likelihood, in Proceedings of the IEEE, Vol. 70, September 1982, pp. 975–989. 66. Nehorai, A. and Paldi, E., Vector-sensor array processing for electromagnetic source localization, IEEE Trans., Vol. SP-42, No. 2, 1994, pp. 376–398. 67. Krolik, J. and Anderson, R., Maximum likelihood coordinate registration for over-the-horizon radar, IEEE Trans., Vol. SP-45, No. 4, 1997, pp. 945–959.

Copyright 2005 by CRC Press

1598_book.fm Page 87 Monday, October 11, 2004 1:57 PM

Correlation Function of Target Return Signal Fluctuations

87

68. Johnson, N., Kotz, S., and Balakrishnan, N., Continuous Univariate Distributions, Vol. 2, John Wiley & Sons, New York, 1995. 69. Zubkovich, S., Statistical Characteristics of Radio Signals Reflected by the Earth Surface, Soviet Radio, Moscow, 1968 (in Russian).

Copyright 2005 by CRC Press

1598_book.fm Page 89 Monday, October 11, 2004 1:57 PM

3 Fluctuations under Scanning of the Three-Dimensional (Space) Target with the Moving Radar

3.1

Slow and Rapid Fluctuations

In the study of target return signal ﬂuctuations caused by the moving radar, we suppose: • The radar antenna is stationary, ∆ϕ sc = ∆ψ sc = 0 .

(3.1)

• Fluctuations caused by the wind are absent, ∆ξ = ∆ζ = 0 .

(3.2)

• Scatterers are in the long range of the radar antenna directional diagram, ∆ϕ rm = ∆ψ rm = 0 .

(3.3)

Let us assume that the radar moves uniformly and linearly with velocity V: ∆ = − V ⋅ τ

and

∆ρ = − Vr ⋅ τ ,

(3.4)

where Vr is the radial component of scatterer velocity relative to the radar. When the radar is brought closer to scatterers, the condition V > 0 is true. Changeover from the space displacement ∆ρ to the shift in time τ implies a changeover from the space ﬂuctuations to the time ﬂuctuations and, consequently, a changeover from the space–time correlation function to the time correlation function.1–3 89 Copyright 2005 by CRC Press

1598_book.fm Page 90 Monday, October 11, 2004 1:57 PM

90 3.1.1

Signal and Image Processing in Navigational Systems General Statements

Using Equation (2.18), Equation (2.74), and Equation (2.92)–Equation (2.95) under the previously mentioned conditions, we can write the total correlation function of target return signal ﬂuctuations in the following form: R(t , τ) = p(t) ⋅ R( τ) = p(t) ⋅ Rp ( τ) ⋅ Rg (τ) ⋅ e jω 0τ ,

(3.5)

where ∞

Rp (τ) =

∑R

en p

(τ − nTp′) ;

(3.6)

n= 0

∫ P(z) ⋅ P (z + µτ) dz ; (τ) = ∫ Π (z) dz ∗

en p

R

(3.7)

2

∫∫ g (ϕ, ψ) ⋅ e dϕ dψ ; R (τ) = ∫∫ g (ϕ, ψ) dϕ dψ jΩ( ϕ , ψ ) τ

2

g

Ω(ϕ , ψ ) =

(3.8)

2

2 Vr ω 0 4 π Vr = = Ω max cos(β 0 + λ c

ϕ cos γ 0

) cos(ψ + γ 0 )

(3.9)

is the Doppler frequency for a scatterer with the coordinates β and γ; Ω max =

2 Vω 0 4 π V = c λ

(3.10)

is the maximum Doppler frequency; µ = 1+

2 Vr0 c

= 1+

2V ⋅ cos β 0 cos γ 0 c

and

Tp′ =

Tp µ

.

(3.11)

Thus, in scanning of the three-dimensional (space) target with pulsed searching signals, the total normalized correlation function is deﬁned as the product of two normalized correlation functions R( τ) = Rp ( τ) ⋅ Rg (τ) ⋅ e jω 0τ .

Copyright 2005 by CRC Press

(3.12)

1598_book.fm Page 91 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Three-Dimensional (Space) Target

91

The ﬁrst normalized correlation function takes into consideration the target return signal ﬂuctuations caused by propagation of periodic pulsed searching signals inside the target, i.e., the ﬂuctuations in the radar range or the rapid target return signal ﬂuctuations. The second normalized correlation function takes into consideration the target return signal ﬂuctuations caused by differences in the radial velocities of scatterers moving relative to the radar antenna, i.e., the Doppler or the slow target return signal ﬂuctuations.4–6 The target return signal deﬁned by the correlation function given in Equation (3.5) is a nonstationary stochastic process, but it is a separable process; the normalized correlation function of target return signal ﬂuctuations determined by Equation (3.12) is independent of the parameter t (time). All information about the nonstationary state is included in the function p(t) given by Equation (2.78), which was studied in more detail in Section 2.4.3. Because of this, our consideration is limited only to the study of the normalized correlation function given by Equation (3.12). With the continuous nonmodulated pulsed searching signal, the condition Rp(τ) 1 is true and the total normalized correlation function of the ﬂuctuations coincides with the normalized correlation function of the slow ﬂuctuations, as the rapid ﬂuctuations are absent. If the radar is stationary, the condition Rg(τ) 1 is true and there are only the rapid (intraperiod) ﬂuctuations under the condition µ = 1. The slow interperiod ﬂuctuations are absent.7–9 Thus, the normalized correlation functions given by Equation (3.6) and Equation (3.8) have a well-founded physical meaning. Let us consider each normalized correlation function individually and, after that, investigate the total normalized correlation function given by Equation (3.12).

3.1.2

The Fluctuations in the Radar Range

The normalized correlation function Rp(τ) given by Equation (3.6) is a periodic function with respect to the variable τ, the period Tp consisting of narrow waves, the width of which is deﬁned by the duration of the function Π(t), i.e., by the pulsed signal duration τp (see Figure 3.1a). The meaning of this dependence is as follows. Under the condition τ < τp , the normalized correlation function Rp(τ) differs from zero because the pulsed signals shifted by τ are partially overlapped. As τ is increased, the pulsed signal overlapping is decreased and a correlation dies out gradually. If the condition τ > τp is true, the pulsed signal overlapping is absent and the normalized correlation function Rp(τ) is equal to zero, i.e., Rp(τ) = 0. In other words, the correlation interval of the target return signal ﬂuctuations in the radar range becomes very close to the pulsed signal duration. With an increase in the value of τ, so long as the value of τ becomes close or equal to the period Tp , a correlation between the pulsed signals appears again, as the pulsed signal of the next period is an exact copy of the pulsed signal in the previous period.10,11 For an unchanged “radar–scatterer” system, in which the radar, radar antenna, and scatterers are stationary, the pulsed searching signals are exact

Copyright 2005 by CRC Press

1598_book.fm Page 92 Monday, October 11, 2004 1:57 PM

92

Signal and Image Processing in Navigational Systems Rp (τ)

τ 0

T ′p Tp

2T ′p 2Tp (a)

Sp (ω) Spen (ω)

ω 0

Ωp Ω′p

(b) FIGURE 3.1 (a) The correlation function and (b) power spectral density of the intraperiod ﬂuctuations.

copies of each other and the correlation between the signals reverts to its original shape completely under the condition τ = nTp , when the same target return signals (the signals received from the same radar range or distance) are compared, i.e., when Rp(nTp) = 1. For this system, the condition Rg(τ) = R∆ξ,∆ζ (τ) 1 is true and correlation properties are completely deﬁned by the normalized correlation function Rp(τ) under the condition µ = 1 (see the solid line in Figure 3.1a). When the radar moves, i.e., when µ ≠ 1, the correlation between the pulsed searching signals becomes maximum under the condition τ = nTp′

or

Rp (nTp′ ) = 1

(3.13)

when the target return signals received from the same pulse volume removed during the time interval nTp′ with the distance determined by Copyright 2005 by CRC Press

1598_book.fm Page 93 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Three-Dimensional (Space) Target

93

Rg (τ)

τ 0

Tp

2Tp (a) Sg (ω)

ω 0 (b) FIGURE 3.2 (a) The correlation function and (b) power spectral density of the interperiod ﬂuctuations.

∆ρ0 = nTp′ V cos β 0 cos γ 0

(3.14)

are compared (see the dotted line in Figure 3.1a). In other words, the radar motion, which is deﬁned by replacing the argument τ with the argument µτ in the normalized correlation function Rp(τ) given by Equation (3.6), implies compression under the condition µ > 1 or expansion under the condition µ < 1 of the time scale µ times. Therefore, the period Tp and the width of the waves are decreased (or are increased) µ times. This is a natural manifestation of the Doppler effect, which is accompanied by changes in pulsed searching signal parameters such as duration, frequency of signal iteration, and changes in the carrier frequency of the signal under radar antenna scanning.12–14 Naturally, when the radar moves, the correlation resumes its original shape incompletely under the condition given by Equation (3.13). Correlation Copyright 2005 by CRC Press

1598_book.fm Page 94 Monday, October 11, 2004 1:57 PM

94

Signal and Image Processing in Navigational Systems

R (τ) = Rp (τ)*Rg (τ)

τ 0

T ′p

2T ′p (a) S(ω) = Sp(ω)*Sg(ω)

ω Ω′p

0

2Ω′p

(b) FIGURE 3.3 (a) The correlation function and (b) power spectral density of the total ﬂuctuations.

broken by the moving radar is taken into consideration by the normalized correlation function Rg(τ) that deﬁnes the target return signal Doppler ﬂuctuations, which are slower ﬂuctuations in comparison with the target return signal ﬂuctuations in the radar range. This normalized correlation function of the Doppler ﬂuctuations will be studied in more detail in the following sections. The power spectral density of intraperiod target return signal ﬂuctuations, which correspond to the periodic normalized correlation function Rp(τ), has the following form: ∞

Sp (ω ) = Spen (ω ) ⋅

∑ n= 0

Copyright 2005 by CRC Press

∞

δ(ω − nΩ′p ) =

∑S

en p

n= 0

(nΩ′p ) ⋅ δ(ω − nΩ′p ) ,

(3.15)

1598_book.fm Page 95 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Three-Dimensional (Space) Target

95

where Ω ′p =

2π = µ Ωp . Tp′

(3.16)

The envelope of the power spectral density of intraperiod ﬂuctuations has the following form [see Equation (2.6)]: p S (ω ) = π en p

∫R

en p

−

(µτ) ⋅ e

− jωτ

jω t

2

p | P (t ) ⋅ e µ dt| dτ = ⋅ ∫ . 2 ∫ Π (t ) dt π

(3.17)

The power spectral density given by Equation (3.15) is a regulated function. This power spectral density is a sequence of discrete harmonics separated from each other by the frequency Ω′p . The envelope of the power spectral density given by Equation (3.17) is the energy spectrum of the pulsed searching signal expanded or compressed µ times (see the dotted line in Figure 3.1b). The shape of the normalized correlation function and the effective power spectral density bandwidth of intraperiod ﬂuctuations are completely deﬁned by the shape and direction of the pulsed searching signal. In the following examples, we consider only the envelope of the normalized correlation function Rpen ( τ) determined by Equation (3.7) and the envelope of the power spectral density Spen (ω ) given by Equation (3.17). The normalized correlation function Rp(τ) and the regulated power spectral density Sp(ω) can be deﬁned without any difﬁculty using Equation (3.6) and Equation (3.15), respectively.15,16 3.1.2.1

The Square Waveform Target Return Signal without Frequency Modulation In this case, we can write Rpen ( τ) = 1 −

|µτ| | τ| , = 1− τp τ ′p

|τ| ≤ τ ′p ;

Spen (ω ) ≈ sinc 2 (0.5ω τ ′p ) , where τ ′p =

τp µ

(3.18)

(3.19)

.

3.1.2.2

The Gaussian Target Return Signal without Frequency Modulation In this case, we can write µτ 2

Rpen ( τ) ≅ e

Copyright 2005 by CRC Press

− 0.5 π ⋅ ( τ p

)

=e

2 − π ⋅ τ2

τc

;

(3.20)

1598_book.fm Page 96 Monday, October 11, 2004 1:57 PM

96

Signal and Image Processing in Navigational Systems

Spen (ω ) ≅ e

2 − π⋅ ω 2

∆Ω p

,

(3.21)

where τ c = 2 τ ′p

2π 2π . = τc τ ′p

∆Ω p = 2 π ∆Fp =

and

(3.22)

Here ∆Fp is the effective power spectral density bandwidth of the ﬂuctuations. 3.1.2.3

The Smoothed Target Return Signal without Frequency Modulation In this case, the shape of the target return signal can be thought as intermediate between the square waveform and the Gaussian shape:

()

Π t =

Φ

(

) − Φ( 2Φ ( )

t + 0.5 τ 0 0.5 τ fr

t − 0.5 τ 0 0.5 τ fr

τ0 τ fr

),

(3.23)

where Φ(x) is the error integral given by Equation (1.27); τfr is the duration of the pulsed searching signal front between levels 0.08 and 0.92, respectively; τ0 is the pulsed searching signal duration at the level 0.5 under the condition τfr ≤ τ0. Equation (3.23) covers, in particular, the cases of the square waveform pulsed searching signal under the condition Gaussian pulsed searching signal under the condition pulsed searching signal duration given by τ ef =

τ0

τ0 τf

r

τ0 τf

r

→ ∞ and the

→ 0. The effective

(3.24)

τ

Φ ( τ f0 ) r

differs from τ0 by not more than 6%. For the limiting case of the Gaussian pulsed signal, we have τef = 0.6τ0. The normalized correlation function of the target return signal ﬂuctuations in the case of the pulsed searching signal with the function Π(t) given by Equation (3.23) is very cumbersome and is omitted here. The envelope of the power spectral density has the following form: S (ω ) ≅ sinc (0.5ω τ ′0 ) ⋅ e en p

Copyright 2005 by CRC Press

2

−

ω τ′

(2

fr

2

)2

.

(3.25)

1598_book.fm Page 97 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Three-Dimensional (Space) Target

97

The envelope of the power spectral density given by Equation (3.25) is the product of the envelope of the power spectral density with the square waveform pulsed searching signal having a duration τ′0 and the envelope of the power spectral density with the Gaussian pulsed searching signal having a duration 0.5 π τ ′fr , where τ ′0 =

τ0 µ

τ ′fr =

and

τ fr µ

.

(3.26)

The effective bandwidth of the power spectral density envelope takes the following form 2

∆Fp = Φ( x) −

1 − e−x 1 , ⋅ π x τ ′0

(3.27)

where x= 2⋅

τ ′0 τ ′fr

(3.28)

(see Figure 3.4). For two limiting cases, as τfr → 0 and τ0 → 0, Equation (3.25) coincides with Equation (3.19) and Equation (3.21). ∆Fp τ ′0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1

1.41 0

τ0 τfr

0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0

FIGURE 3.4 The normalized effective bandwidth of envelope of the power spectral density of the ﬂuctuations in the radar range with the smoothed pulsed searching signal.

Copyright 2005 by CRC Press

1598_book.fm Page 98 Monday, October 11, 2004 1:57 PM

98

Signal and Image Processing in Navigational Systems

3.1.2.4

The Square Waveform Target Return Signal with LinearFrequency Modulation In this case, we can write: ω (t ) = ω 0 +

∆ω M ⋅t ; τp

t

Ψ(t) =

∆ω M

∫ (ω − ω ) dt = 2 τ 0

0

(3.29)

⋅ t2 ,

(3.30)

p

where ∆ωM is the deviation in frequency during the time interval Tp. Therefore, 2 τ ⋅ sin[0.5∆ω ′M τ(1 − || τ ′p )] ; ∆ω ′M τ

(3.31)

p ⋅ {[C( x 2 ) − C( x1 )] 2 + [S( x 2 ) − S( x1 )] 2 } , ∆ω ′M

(3.32)

Rpen ( τ) =

Spen (ω ) ≅ where

x 1, 2 =

[ 2(ω∆ω−′ ω) ∓ 1]⋅ 0

0.25π D ;

(3.33)

M

∆ω ′M = µ ∆ω M ; D=

∆ω ′M τ ′p 2π

= ∆ fM ⋅ τ p

(3.34)

(3.35)

is the relative deviation in frequency; C(x) and S(x) are the Fresnel integrals. The normalized correlation function Rpen ( τ) determined by Equation (3.31) (see Figure 3.5) tends to approach the normalized correlation function given by Equation (3.18) as D → 0. This is true when frequency modulation is absent. At D >> 1 and τ << τp , this function becomes close to the function sinc2(0.5∆ω′M τ), shown in Figure 3.5 by the dotted line. Under changing of the relative deviation in frequency, the normalized correlation function and the envelope of the power spectral density are deformed. At low values of D, the envelope of the power spectral density is approximately close to one for the case of the square waveform pulsed searching signal given by Equation (3.19), with the effective bandwidth equal to (τ′p )–1. With high frequency deviation, the envelope of the power spectral density tends to approach the Copyright 2005 by CRC Press

1598_book.fm Page 99 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Three-Dimensional (Space) Target

99

Rpen (τ) 1.0 0.9 0.8 1

0.7 0.6 0.5

2

5

3 4

6

0.4 0.3 0.2 0.1

τ τ′p

7

0

0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

−0.1 −0.2 FIGURE 3.5 The normalized correlation function of the ﬂuctuations in the radar range with the square waveform linear-frequency modulated pulsed searching signal, n = 0: (1) D = 0; (2) D = 0.5; (3) D = 1; (4) D = 2; (5) D = 4; (6) D = 10; (7) D = 20.

square waveform, one with the bandwidth equal to deviation in frequency ∆f ′M during the pulsed searching signal duration (see Figure 3.6). In the general case, the correlation interval of the target return signal, which is inversely proportional to the effective bandwidth of the power spectral density envelope, has the following form: 2 τc = ⋅ [C 2 (0.5 τp D

π D )]

(3.36)

∆ fM → 0 .

(3.37)

π D ) + S2 (0.5

and is shown in Figure 3.7. We see that τ c → τ ′p

as

D→0

or

This statement can be proved using the L’Hospital rule twice in Equation (3.36).17,18 As D → ∞ or as ∆fM → ∞, the correlation interval tends to approach τc → (∆f′M )–1 and is deﬁned by the deviation in frequency during the pulsed searching signal duration. Because the correlation interval of the target return signal is equal to the signal duration when frequency modulation is absent, we can say that the linear-frequency modulation from the standpoint of the Copyright 2005 by CRC Press

1598_book.fm Page 100 Monday, October 11, 2004 1:57 PM

100

Signal and Image Processing in Navigational Systems

Spen (ω)

fM τp −80

−60

−40

−20

0

20

40

60

80

FIGURE 3.6 The envelope of the power spectral density of the ﬂuctuations in the radar range with the square waveform linear-frequency modulated pulsed searching signal, D = 100.

τc τc |D =100 1.0 0.9 0.8 1 0.7 0.6 2

0.5 0.4 0.3 0.2 0.1

D 0

5

10

15

20

FIGURE 3.7 The normalized correlation interval of the ﬂuctuations in the radar range as a function of the parameter D = ∆fM · τp: (1) the linear-frequency modulated pulsed searching signal; (2) the Gaussian pulsed searching signal.

Copyright 2005 by CRC Press

1598_book.fm Page 101 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Three-Dimensional (Space) Target

101

correlation interval of the target return signal is equivalent to the truncation of the pulsed searching signal D times, i.e., prior to the duration (∆f′M )–1. This is to ﬁt the well-known phenomenon of improvement in the radar range resolution with high-frequency modulation. 3.1.2.5

The Gaussian Target Return Signal with Linear-Frequency Modulation In this case, the normalized correlation function and the envelope of the power spectral density of target return signal ﬂuctuations are determined by Equation (3.20) and Equation (3.21), in which τc =

2 τ ′p

,

1 + ∆ fM2 τ 2p

(3.38)

where ∆ fM τ p = ∆ fM′ τ ′p .

(3.39)

If ∆ fM τ p << 1 ,

τ c = 2 τ ′p

(3.40)

τ c = ( ∆ fM′ ) −1 .

(3.41)

then

and if ∆ fM τ p >> 1 ,

then

The function τc 2 τ ′p

=

1 1 + D2

(3.42)

given by Equation (3.38)–Equation (3.41) is shown in Figure 3.7 by the dotted line.

3.1.3

The Doppler Fluctuations

The target return signal Doppler ﬂuctuations are deﬁned by the normalized correlation function Rg(τ) given by Equation (3.8). This function Rg(τ) is slow in comparison with the normalized correlation function Rpen ( τ) of the ﬂuctuations given by Equation (3.7).19,20 The function Rg(τ) is the envelope of the

Copyright 2005 by CRC Press

1598_book.fm Page 102 Monday, October 11, 2004 1:57 PM

102

Signal and Image Processing in Navigational Systems

normalized correlation function R(τ) given by Equation (3.12) and characterizes the interperiod ﬂuctuations (see Figure 3.2a). When the target return signal at the receiver or detector input in navigational systems is a continuous, nonmodulated stochastic process, if the ﬂuctuations in the radar range are absent, the normalized correlation function Rg(τ) deﬁnes the ﬂuctuations completely.21,22 Let us deﬁne the power spectral density of the Doppler ﬂuctuations corresponding to the normalized correlation function Rg(τ). Multiplying the normalized correlation function Rg(τ) by the factor ejω0τ and using the Fourier transform, we can write

Sg (ω ) ≅

∫∫

∞

g 2 (ϕ , ψ ) dϕ dψ

∫e

j[ ω 0 + Ω ( ϕ , ψ ) − ω ] τ

dτ ≅

−∞

∫∫ g (ϕ, ψ) 2

(3.43)

× δ ω 0 + Ω(ϕ , ψ ) − ω  dϕ dψ Using the ﬁltering properties of the delta function, the double integral in Equation (3.43) can be reduced to a simple integral. For this purpose, the condition Ω(ϕ , ψ ) = Ω max cos(β 0 +

ϕ cos γ 0

) cos(ψ + γ 0 ) = ω − ω 0

(3.44)

must be satisﬁed [see Equation (3.9)]. In the general case, determination of the integral in Equation (3.43) could require the use of numerical techniques. In the majority of important cases in practice, we can determine the integral in Equation (3.43) without using numerical techniques, which is very important.23,24 Equation (3.44) is equivalent to the formula cos θ = (ω − ω 0 ) ⋅ (Ω m ) −1 = ν ,

(3.45)

where θ is the angle between the vector of a velocity of a moving aircraft (radar), for example, and the direction of radar antenna scanning [see Equation (2.91) and Equation (2.92)]. This means that the delta function in Equation (3.43) differs from zero when the angles ϕ and ψ satisfy the following condition: θ = arccos ν = const. In other words, the power spectral density at the relative Doppler frequency is formed by summing the powers of the target return signals from those scatterers that are placed on the surface of the cone deﬁned near the velocity vector of a moving aircraft (radar) with apex angle 2θ. This geometric representation of the formation of the Doppler ﬂuctuations with moving radar can be used for determination of the power spectral density of Doppler ﬂuctuations.25,26

Copyright 2005 by CRC Press

1598_book.fm Page 103 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Three-Dimensional (Space) Target

103

Assuming that θ ∈ [0, π] in Equation (3.45), we can deﬁne that the spectrum given by Equation (3.44) is always within the limits of the interval ω 0 − Ω max ≤ ω ≤ ω 0 + Ω max .

3.2

(3.46)

The Doppler Fluctuations of a High-Deflected Radar Antenna

3.2.1

The Power Spectral Density for an Arbitrary Directional Diagram

Equation (3.43) can be fundamentally simpliﬁed and easily studied if we assume that the radar antenna directional diagram axis is deﬂected from the direction of moving radar so that at least one of the angles β0 or γ0 is greater than the directional diagram width in the corresponding plane. In this case, reasoning that the width is small, we can use the Taylor-series expansion27,28 for the function Ω(ϕ, ψ) in Equation (3.9), limiting terms to the ﬁrst order: Ω ′(ϕ , ψ ) ≅ Ω 0 − ϕ Ω h − ψ Ω v ,

(3.47)

Ω 0 = Ω max cos β 0 cos γ 0 ;

(3.48)

2 ω 0V 4 π V = ; c λ

(3.49)

where

Ω max =

Ω h = Ω max sin β 0 ;

(3.50)

Ω v = Ω max cos β 0 sin γ 0 .

(3.51)

Here, Ω0 is the Doppler frequency corresponding to the center of the pulse volume — a direction along the directional diagram axis. At ﬁrst, we assume that the variables ϕ and ψ in the function g(ϕ, ψ) are separable. Substituting Equation (3.47) in Equation (3.8), we can write Rg (τ) = Rgh (τ) ⋅ Rgv (τ) ⋅ e j(ω 0 + Ω0 ) τ , where

Copyright 2005 by CRC Press

(3.52)

1598_book.fm Page 104 Monday, October 11, 2004 1:57 PM

104

Signal and Image Processing in Navigational Systems

∫ g (ϕ) ⋅ e dϕ R ( τ) = ∫ g (ϕ) dϕ

(3.53)

∫ g ( ψ ) ⋅ e dψ R ( τ) = ∫ g (ψ) dψ

(3.54)

h g

− jϕ Ω h τ

2 h

2 h

and v g

− jψ Ω v τ

2 v

2 v

are the normalized correlation functions with the corresponding power spectral densities Sgh (ω ) ≅ g h2 ( − Ωωh )

and

Sgv (ω ) ≅ g v2 ( − Ωωv ) .

(3.55)

This is obvious, as Equation (3.53) and Equation (3.54) are the Fourier transforms of the power spectral densities given by Equation (3.55). The power spectral densities of target return signal Doppler ﬂuctuations given by Equation (3.55) coincide in shape with the square of the directional diagram in the horizontal and vertical planes. The total power spectral density corresponding to the normalized correlation function given by Equation (3.52) is deﬁned by the convolution of the power spectral densities determined by Equation (3.55) and a shift of convolution by ω0 + Ω0: Sg (ω ) = Sgh (ω ) ∗ Sgv (ω ) ∗ δ(ω − ω 0 − Ω0 ) ≅

∫ g (− 2 h

x Ωh

) ⋅ gv2 (

ω 0 + Ω0 − ω + x Ωv

) dx (3.56)

If both angles β0 and γ0 are very small, the formulae obtained based on Equation (3.47) are not true. If β0 = γ0 = 0, the equality Rgh ( τ) = Rgv ( τ) ≡ 1 follows from these formulae. This means that the Doppler ﬂuctuations are absent. However, this statement, rigorously speaking, is not true because under the conditions β0 → 0 and γ0 → 0, the effective power spectral density bandwidth of the Doppler ﬂuctuations is decreased straight away but does not equal zero. This will be proved in Section 3.3. If one of the angles β0 and γ0 is equal to zero, for example, if γ0 = 0, we can write that Rgv ( τ) ≡ 1 and Sgv (ω ) = δ(ω ). Thus, the power spectral density Sg(ω) given by Equation (3.56) coincides with the power spectral density Sgh (ω ) shifted by ω 0 + Ω 0 = ω 0 + Ω max cos β 0 . So, we can write

Copyright 2005 by CRC Press

(3.57)

1598_book.fm Page 105 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Three-Dimensional (Space) Target

Sg (ω ) = Sgh (ω − ω 0 − Ω0 ) ≈ gh2 (

ω 0 + Ω0 − ω Ωh

).

105

(3.58)

In an analogous way, under the condition β0 = 0, we can write Rgh ( τ) ≡ 1, Sgh (ω ) = δ(ω ), and Sg (ω ) = Sgv (ω − ω 0 − Ω0 ) ≈ gv2 (

ω 0 + Ω0 − ω Ωv

),

(3.59)

where Ω 0 = Ω max cos γ 0

(3.60)

Ω v = Ω max sin γ 0 .

(3.61)

and

In spite of the fact that Equation (3.58) and Equation (3.59) are approximate, as rigorously speaking, Sgv (ω ) ≠ δ(ω ) at γ0 = 0 and Sgh (ω ) ≠ δ(ω ) at β0 = 0, they can give us sufﬁcient accuracy in the majority of cases. An exception to this rule is the case where ∆h << ∆v at γ0 = 0 and ∆v << ∆h at β0 = 0. Let us continue to consider the general case when the variables ϕ and ψ in the directional diagram g(ϕ, ψ) need not be separable. Substituting Equation (3.47) in Equation (3.43), we can write S g (ω ) ≈

∫ g (ϕ, ψ) ⋅ δ(ω 2

0

+ Ω 0 − ϕ Ω h − ψ Ω v − ω ) dϕ dψ .

(3.62)

Using the ﬁltering property of the delta function,29,30 we can carry out integration with respect to only one variable, for example, ϕ: Sg (ω ) ≈

∫g ( 2

ω 0 + Ω0 − ω − ψ Ωv Ωh

, ψ ) dψ .

(3.63)

If the variables ϕ and ψ are separable, Equation (3.56) follows from Equation (3.63). Let us introduce a new system of coordinates ϕ′ and ψ′ rotated by some angle κ relative to the coordinate system ϕ and ψ: ϕ = ϕ ′ cos κ − ψ ′ sin κ ;

(3.64)

ψ = ϕ ′ sin κ + ψ ′ cos κ .

(3.65)

Choose the angle κ so that the conditions

Copyright 2005 by CRC Press

1598_book.fm Page 106 Monday, October 11, 2004 1:57 PM

106

Signal and Image Processing in Navigational Systems

cos κ =

Ωh G

and

sin κ =

Ωv G

(3.66)

are satisﬁed, where G = Ω 2h + Ω 2v = Ω max sin 2 β 0 + cos 2 β 0 sin 2 γ 0 .

(3.67)

Then, instead of Equation (3.62), we can write S g (ω ) ≈

∫∫ g (ϕ′, ψ ′ ) ⋅ δ(ω 2

0

+ Ω 0 − ω − Gϕ ′ ) dϕ ′ dψ ′ ,

(3.68)

where g ( ϕ ′ , ψ ′ ) = g( ϕ , ψ )

(3.69)

is the directional diagram in the coordinate system ϕ ′ and ψ ′ , or S g (ω ) ≈

∫g ( 2

ω 0 + Ω0 −ω Gψ ′

) dψ ′ .

(3.70)

The condition in Equation (3.66) means that we must choose the axes ϕ′ and ψ′ in such a manner that the plane ϕ′(ψ′ = 0) coincides with the plane crossing the velocity vector of moving aircraft (radar) and the directional diagram axis. It is possible to show that sin 2 β 0 + cos 2 β 0 sin 2 γ 0 = sin θ 0

and

cosβ 0 cos γ 0 = cos θ 0 , (3.71)

where θ0 is the angle between the directional diagram axis and the direction of moving radar. Because of this, we use the following forms in Equation (3.48), Equation (3.49), and Equation (3.70): G = Ω max sin θ 0

and

Ω 0 = Ω max cos θ 0 .

(3.72)

To make clear the physical meaning of Equation (3.70), we assume g (ϕ ′ , ψ ′ ) = g h (ϕ ′ ) ⋅ g v ( ψ ′ ) ,

(3.73)

which is analogous to Equation (2.97). Then Equation (3.70) can be represented in the simpler form:31

Copyright 2005 by CRC Press

1598_book.fm Page 107 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Three-Dimensional (Space) Target ω +Ω −ω

0 0 Sg (ω ) ≈ gh2 ( Ωmax ). sin θ0

107

(3.74)

Thus, if the radar antenna axis is deﬂected from the direction of moving radar and the condition given by Equation (3.73) is satisﬁed, then the power spectral density of the Doppler ﬂuctuations coincides in shape with the square of the directional diagram in the plane crossing the direction of radar motion and the directional diagram axis. The frequency corresponding to the maximum power spectral density has the following form: ω = ω 0 + Ω 0 = ω 0 + Ω max cos θ 0 = ω 0 +

4πV ⋅ cos β 0 cos γ 0 . λ

(3.75)

The power spectral density bandwidth given by Equation (3.74) is deﬁned by the width ∆˜ (a2 ) of the square of the directional diagram by power in the plane of deﬂection: ∆F =

2 V ˜ (2) ⋅ ∆ a sin θ 0 . λ

(3.76)

Thus, ∆F can be considered as the effective power spectral density bandwidth of the Doppler ﬂuctuations and ∆˜ (a2 ) can be considered as the directional diagram width at some arbitrary level. If ∆˜ (a2 ) denotes the effective width, then we can write ∆˜ (a2 ) = k a ∆˜ a , where ka is a coefﬁcient deﬁning the shape of the directional diagram [see Equation (2.99)]. As the directional diagram width has the form ∆ a = εdaλ , where ε is a coefﬁcient that is approximately equal to the unit and da is the radar antenna diameter, reference to Equation (3.76) allows us to arrive at a very important conclusion: With the deﬂected directional diagram, the effective power spectral density bandwidth of the Doppler ﬂuctuations is independent of the wavelength λ. This can be explained as follows. With an increase (decrease) in the wavelength λ, the width ∆a of the directional diagram increases (decreases) and the Doppler frequency 2λV decreases (increases) simultaneously. The functions of the effective bandwidth ∆Ω = 2π∆F of the power spectral density of the Doppler ﬂuctuations and frequency Ω0 versus the angle θ0 are shown in Figure 3.8 by the solid and dotted lines, respectively. Reference to Equation (3.76) shows that the space correlation interval has the following form: ∆ c = V ⋅ τ c =

Copyright 2005 by CRC Press

da da V = ≈ . ∆F 2 k a ε sin θ 0 sin θ 0

(3.77)

1598_book.fm Page 108 Monday, October 11, 2004 1:57 PM

108

Signal and Image Processing in Navigational Systems

∆Ωmax = Ωmax ∆α(2) ∆Ω

Ω0 θ0

−Ω max

V Ω0

Ω max = 4 πV λ

∆Ωmax FIGURE 3.8 The bandwidth ∆Ω and average frequency Ω0 of the power spectral density of the Doppler ﬂuctuations as a function of the radar antenna deﬂection angle θ0.

Hence, it follows that if θ = 90°, then ∆c ≈ da , i.e., with the deﬂection of a radar antenna by da , the target return signals become uncorrelated.32 With a decrease in the angle θ0 , the correlation interval increases (see Figure 3.9) and tends to approach ∞ as θ0 → 0, but the last statement is not true because under the condition θ0 = 0, Equation (3.76), rigorously speaking, is not true (see Section 3.3). If the condition given by Equation (3.73) is not satisﬁed, then the power spectral density of the Doppler ﬂuctuations is determined by Equation (3.70). The meaning of the formula in Equation (3.70) is the same as the meaning of the formula in Equation (3.74). A difference is that according to the deﬁnition of the power spectral density at some frequency ω, it is necessary to carry out integration with respect to the variable ψ′ in Equation (3.70). The result of this integration differs at various frequencies because the condition given by Equation (3.73) is not satisﬁed. For this case, rigorously speaking, we cannot state that the shape of the power spectral density coincides with the shape of the radar antenna directional diagram in the plane of radar antenna deﬂection, i.e., in the plane ψ′ = 0, because for various values of the variable ψ′, the directional diagrams are different. However, integration with respect to the variable ψ′ means averaging with respect to the variable ψ′ and, in practice, the right side of Equation (3.70) is the average directional diagram in the plane of radar antenna deﬂection. In particular, if the directional diagram is symmetrical

Copyright 2005 by CRC Press

1598_book.fm Page 109 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Three-Dimensional (Space) Target

109

θ0 = π 2

∆c ≅ da (a)

θ0

∆c ≅

da sin θ0

(b) FIGURE 3.9 The space correlation interval ∆c as a function of the angle θ0: (a) ∆ c ≅ d a ; (b) ∆ c ≅

da sin θ 0

.

with respect to the axis, then the shape of the power spectral density is independent of the plane in which the directional diagram is deﬂected.33,34 It is not difﬁcult to explain from the physical viewpoint why the power spectral density coincides in shape with the square of the directional diagram with moving radar. The frequency ω of the target return signal from some elementary scatterers is different from the frequency ω0 by the Doppler frequency Ω = ω − ω 0 = Ω max cos θ = Ω max cos β cos γ ,

(3.78)

where ω0 is the carrier frequency of the pulsed searching signal and θ is the angle between the direction of moving radar and that toward the given elementary scatterers.

Copyright 2005 by CRC Press

1598_book.fm Page 110 Monday, October 11, 2004 1:57 PM

110

Signal and Image Processing in Navigational Systems

ρ

θ

cτp 2 FIGURE 3.10 Formation of the signal with Doppler frequency Ω = Ωmax cos θ.

Obviously, the same Doppler frequency corresponds to all scatterers in which θ = const, i.e., to all scatterers placed on the surface of the cone whose axis coincides with the direction of moving radar and whose apex angle is 2θ. Scatterers placed on the surface of the corresponding cone, which are also in the pulse volume (i.e., are placed at a distance that is within the limits of the interval [ρ – 0.25 cτp , ρ + 0.25 cτp] from the radar), take part in generation of the target return signals with the given Doppler frequency received at some instant of time (see Figure 3.10). Both the average power and the average power spectral density of the ﬂuctuations with the given Doppler frequency are equal to the sum of powers of elementary signals with the same Doppler frequency.35,36 Figure 3.11 shows a sphere with radius ρ — the radar antenna is placed at the center — containing the circle C that is formed as a result of intersecting the sphere with a cone having apex angle 2θ. Summing the powers of elementary signals from scatterers placed on the segment C′ of the circle C within the limits of the region covered by the directional diagram (the hatched region), we obtain the power spectral density at the frequency ω0 + Ω, where Ω is determined by Equation (3.78). Evidently, if a deﬂection of the directional diagram axis OA from the direction of moving radar OD is sufﬁciently large, so that the segment C′ does not differ greatly from the straight line within the limits of the directional diagram (which is assumed to be narrow), then the power spectral density depends on the squared ampliﬁcation coefﬁcient along the direction OB as a function of the parameter. The total power spectral density coincides in shape with the square of the directional diagram by power within the cross section B′B″. If the radar antenna directional diagrams are the same at the planes that are perpendicular to the cross section B′B″ — i.e., the condition given by Equation

Copyright 2005 by CRC Press

1598_book.fm Page 111 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Three-Dimensional (Space) Target

B ′′

111

C′ B B′ D

θ1 θ0

C

θ ρ 0

FIGURE 3.11 Formation of the power spectral density of the Doppler ﬂuctuations.

(3.73) is true — then the results of integration over various circles at different values of the angle θ or frequency Ω are the same and do not impact the shape of the power spectral density. When these directional diagrams are not the same, then the power spectral density is different from the square of the radar antenna directional diagram at the cross section B′B″. The power spectral density will coincide with the square of the directional diagram averaged over all cross sections that are parallel to the cross section B′B″. In practice, the real directional diagram consists of the main lobe and side lobes.37,38 The side lobes of the radar antenna directional diagram form remainders of the power spectral density. These remainders must be taken into consideration when solving many problems that arise in practice.

3.2.2

The Power Spectral Density for the Gaussian Directional Diagram

When the condition given by Equation (3.47) is satisﬁed and the radar antenna directional diagram is Gaussian, reference to Equation (3.52) and Equation (3.56) shows that Rg ( τ ) = e

2 −π⋅ τ

Sg (ω ) ≅ e

Copyright 2005 by CRC Press

τ 2c

−π⋅

⋅ e j ( ω 0 + Ω0 ) τ ; ( ω − ω 0 − Ω0 )2 4 π 2 ∆F 2

,

(3.79)

(3.80)

1598_book.fm Page 112 Monday, October 11, 2004 1:57 PM

112

Signal and Image Processing in Navigational Systems

where ∆F =

1 V 2 (2) = ⋅ 2( ∆ h sin 2 β 0 + ∆ (v ) cos 2 β 0 sin 2 γ 0 ) . τc λ

(3.81)

If the condition ∆h = ∆v = ∆a is true, then using Equation (3.71) and Equation (3.81), we can write ∆F =

V∆ a V∆ a ⋅ 2(sin 2 β 0 + cos 2 β 0 sin 2 γ 0 ) = ⋅ 2 sin θ 0 . λ λ

(3.82)

Formulae in Equation (3.80)–Equation (3.82) can be obtained also by convolution of the power spectral density of the Doppler ﬂuctuations given by Equation (3.55) if the condition

Sgh ,v (ω ) ≅ e

− π⋅

ω2 4 π 2 ∆F 2 h ,v

(3.83)

is true, where ∆Fh =

V∆ h ⋅ 2 sin β 0 λ

and

∆Fv =

V∆ v ⋅ 2 cos β 0 sin γ 0 . λ

(3.84)

Comparing Equation (3.81) and Equation (3.84), one can see that ∆F = ∆Fh2 + ∆Fv2 .

(3.85)

The following example can help us to estimate the effective power spectral m density bandwidth: at V = 300 sec , λ = 3 cm, ∆h = ∆v = 2°, and β0 = γ0 = 30°, we obtain ∆F = 330 Hz.

3.2.3

The Power Spectral Density for the Sinc-Directional Diagram

In particular cases that arise in practice, when the radar antenna directional diagram is deﬂected so that the velocity vector of moving radar is in the plane of one of the main cross sections of the directional diagram, and the condition given by Equation (3.73) is satisﬁed, there is no need to deﬁne the convolution of Equation (3.56), and the power spectral density of the Doppler ﬂuctuations can be easily determined using Equation (2.103), Equation (3.53), Equation (3.54), Equation (3.58), and Equation (3.59). For example, under the conditions β0 ≠ 0 and γ0 = 0 for the case of the sinc-diagram, we can write

Copyright 2005 by CRC Press

1598_book.fm Page 113 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Three-Dimensional (Space) Target Sg (ω ) = Sgh (ω ) ≈ sinc 4 (

ω −ω 0 − Ω0 3 ∆Fh

),

113

(3.86)

where Ω0 = Ωmax cos β0 and ∆Fh =

4 V∆ h ⋅ sin β 0 3λ

(3.87)

is the effective power spectral density bandwidth. Comparing Equation (3.84) and Equation (3.87), one can see that with the same effective bandwidth of the directional diagram, the effective bandwidth of the Gaussian power spectral density is 6% more than the effective power spectral density bandwidth given by Equation (3.86). For the case considered, the normalized correlation function has the following form:  1 − (1.5) 3 τ 2 + (1.5) 4 τ 3   Rg ( τ) =  2[1 − 0.75 τ]3   0

at

τ ≤ 0.67 ;

at

0.67 < τ ≤ 1.33;

at

τ > 1.33,

(3.88)

where τ=

| τ| ; τ ch

τ ch = ( ∆Fh ) −1

(3.89)

(3.90)

is the correlation interval. The power spectral density given by Equation (3.86) is shown in Figure 3.12 (the solid line). The normalized correlation function Rh(τ) is shown in Figure 3.13 (the solid line). The normalized correlation function Rh(τ) is deﬁned within the limits of the intervals τ < 0.67 and τ > 0.67 by various functions that have the same derivative at the point τ = 0.67. In the case of the Gaussian directional diagram, the power spectral density (see the dotted line in Figure 3.12) lacks side lobes and is very close to the power spectral density Sg(ω) given by Equation (3.86), within the main lobe. The normalized correlation functions are very close to each other. However, there is an essential difference between them.39,40 The normalized correlation function Rh(τ) given by Equation (3.88) is different from zero within the limits of the ﬁnite interval and the normalized correlation function Rg(τ) given by Equation (3.79) is different from zero within the limits of the inﬁnite interval. This fact gives rise to a difference Copyright 2005 by CRC Press

1598_book.fm Page 114 Monday, October 11, 2004 1:57 PM

114

Signal and Image Processing in Navigational Systems

h,v Sg (ω)

100

10−1

2

10−2

1 10−3

10−4 ω − ω0 − Ω0 10−5 −5 −4

2 π∆ Fh,v −3

−2

−1

0

1

2

3

4

5

FIGURE 3.12 The power spectral density of the Doppler ﬂuctuations at β0 = 0 or γ0 = 0: (1) the sinc-diagram; (2) the Gaussian directional diagram.

Rg (τ) 1.0 0.9 0.8

1

0.7 0.6 2 0.5 0.4 0.3 0.2 τ τc

0.1 0

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3

FIGURE 3.13 The normalized correlation function at β0 = 0 or γ0 = 0: (1) the sinc-diagram; (2) the Gaussian directional diagram.

Copyright 2005 by CRC Press

1598_book.fm Page 115 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Three-Dimensional (Space) Target

115

in the power spectral density shown in Figure 3.12 — the power spectral density given by Equation (3.86) has side lobes. In the general case, i.e., if the conditions β0 ≠ 0 and γ0 = 0 are satisﬁed, the total normalized correlation function Rg(τ) given by Equation (3.52) is deﬁned by the product of two normalized correlation functions Rgh ( τ) and Rgv ( τ) determined by Equation (3.88). In this case, a deﬁnition of the power spectral density in the form of the Fourier transform of the total normalized correlation function Rg(τ) [see Equation (3.52)] or convolution [see Equation (3.56)] of the power spectral densities [see Equation (3.86)] under the condition ∆Fh ≠ ∆Fv is very cumbersome.41 In accordance with Equation (2.103) and Equation (3.55), we can write Sgh (ω ) ≅ sinc 4 ( 3 ω∆Fh ) and Sgv (ω ) ≅ sinc 4 ( 3 ω∆Fv ) ,

(3.91)

where 4 V∆ h ⋅ sin β 0 ; 3λ

(3.92)

4 V∆ v ⋅ cos β 0 sin γ 0 . 3λ

(3.93)

∆Fh =

∆Fv =

Under the condition ∆Fh = ∆Fv = ∆F, which occurs if the condition tg β 0 =

∆v ⋅ sin γ 0 ∆h

(3.94)

is satisﬁed, the convolution [see Equation (3.56)] of the power spectral densities given by Equation (3.91) gives us the following result: Sg (ω ) ≅

1  3+ ν4 

90 ν2

− ( 12ν +

120 ν3

)sin ν − (1 −

60 ν2

) cos ν −

15 ν3

sin 2 ν  , 

(3.95)

where ν=

2(ω − ω 0 − Ω 0 ) . 3 ∆F

(3.96)

The power spectral density Sg(ω) given by Equation (3.95) is shown in Figure 3.14 by the solid line. If the condition ∆Fh ≠ ∆Fv is true, but the difference between ∆Fh and ∆Fv is not so high, we can use Equation (3.95) under the condition

Copyright 2005 by CRC Press

1598_book.fm Page 116 Monday, October 11, 2004 1:57 PM

116

Signal and Image Processing in Navigational Systems

Sg (ω) 100

10−1

1

10−2

10−3

2 3

10−4 1.5 10−5

0

ω − ω0 − Ω0 ∆Ωh,v

0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 5.5 6.0

FIGURE 3.14 The exact and approximate power spectral densities of the Doppler ﬂuctuations for sincdiagram at β0 = 0 or γ0 = 0.

∆F = 0.5( ∆Fh2 + ∆Fv2 ) .

(3.97)

When the difference between ∆Fh and ∆Fv is very high, we can use approximate techniques to deﬁne a convolution of the power spectral densities in addition to computer modeling and computer calculation of integrals.42

3.2.4

The Power Spectral Density for Other Forms of the Directional Diagram

The approximate technique is based on representation of the squares of the radar antenna directional diagram g h2 (ϕ) and g v2 ( ψ ) and the power spectral densities Sgh (ω ) and Sgv (ω ) in the form of the sum of the main lobe (containing the main part of energy and obeying the Gaussian law) and of the remainders, taking into consideration the average side lobes:43 Sgh (ω ) = Shen (ω ) + sh (ω )

and

Sgv (ω ) = Sven (ω ) + sv (ω ) .

(3.98)

Because the energy contained in the power spectral densities sh(ω) and sv (ω) is low, and the length of the power spectral densities sh(ω) and sv (ω) is large

Copyright 2005 by CRC Press

1598_book.fm Page 117 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Three-Dimensional (Space) Target

117

in comparison with that of the main lobe, the convolution has the following form: Sgh (ω ) ∗ Sgv (ω ) ≅ Shen (ω ) ∗ Sven (ω ) + Shen (ω ) ∗ sv (ω ) + Sven (ω ) ∗ sh (ω ) ≈

∫

∫

∫

Shen ( x) ⋅ Sven (ω − x) dx + sv (ω ) Shen ( x) dx + sh (ω ) Sven ( x) dx

. (3.99)

As the power spectral densities Shen (ω ) and Sven (ω ) are Gaussian, we can use the computer to calculate the formula in Equation (3.99) without any difﬁculty. In many practical cases, the squared directional diagram with average near side lobes can be written in the following form44

gh2,v (ϕ) = e

− 2π

(

ϕ

)

2

ε ∆ en h ,v

+

3 8 a2

⋅(

∆ en h ,v n ϕ

)

⋅ ϕ (bε ∆ en h ,v ) ,

(3.100)

where a, n, and b are coefﬁcients characterizing energy or power of sidelobes and the rate of decrease of the side-lobes and depending on the distribution of current along the radar antenna diameter; ∆enh ,v =

λ dh , v

;

ε ∆enh ,v = ∆ h ,v is the effective width of the directional diagram by power; dh,v is the radar antenna diameter in the horizontal and vertical planes, respectively; ε is a coefﬁcient ensuring equality between the width of the real squared directional diagram and the approximated Gaussian function at the level 0.5;  0  ϕ (bε ∆ en ) =  h ,v  1

at

|ϕ| < bε ∆ en h ,v ;

at

|ϕ| > bε ∆ hen,v .

(3.101)

Usually, in the determination of the power spectral density of the Doppler ﬂuctuations, the shape of side lobes of the directional diagram need not be taken into consideration as they are highly smoothed in the formation of the convolution of the power spectral densities. In Table 3.1 the reader can ﬁnd the exact formulae for the three forms of the directional diagrams used in practice and the corresponding power spectral densities obtained on the basis of Equation (3.55). In addition, approximate values of the parameters ε, a, n, and b are presented in Table 3.1. Substituting Equation (3.100) in Equation (3.55), we can deﬁne the power spectral densities S (ω ) ≅ e h ,v g

Copyright 2005 by CRC Press

−π⋅

ω2 ∆Ω2 h ,v

+ G(

2 ∆Ωh , v n ω

)

⋅ ω ( 2 b∆Ωh ,v ) ,

(3.102)

1598_book.fm Page 118 Monday, October 11, 2004 1:57 PM

118

Signal and Image Processing in Navigational Systems

TABLE 3.1 Radar Antenna Directional Diagrams and the Corresponding Spectral Power Densities Distribution Law Uniform

gh,v(ϕ) sinc 2 ( ∆πenϕ

h ,v

cos(

π ∆ c da

cos 2 (

)

π ∆ c da

)

ε

a

n

0.96

10

1.28

1.55

Sh,v(ω) sinc 4 ( π∆Ωω )

)

 cos( πϕ ) ∆en h ,v   2 ϕ  1 − ( en ) 2  ∆   h ,v

2

 sinc( πϕ ) ∆en h ,v   ϕ  1 − ( en ) 2  ∆ h ,v  

2

πω   cos( ∆Ω )  2ω 2  1 − ( )   ∆Ω

4

πω   sinc( ∆Ω )  ω 2  1 − ( )   ∆Ω

4

b

G

4

1.1

4·10–3

16

8

1.4

2·10–4

10

12

1.5

2·10–5

where ∆Ω h ,v = 2 π ∆Fh ,v

(3.103)

and ∆Fh,v is determined by Equation (3.84), in which ∆ h ,v =

ελ dh , v

and

G=

3 , 8 a2ε n

(3.104)

(see Table 3.1). Using the formula in Equation (3.99), we can write

Sg (ω ) ≅

e

2 −π⋅ ω

∆Ω2

Ω

[

+ 22n G

∆Ωnh − 1 ωn

⋅ ω ( 2 b∆Ωh ) +

∆Ωnv − 1 ωn

],

⋅ ω ( 2 b∆Ωv )

(3.105) where ∆Ω = ∆Ω 2h + ∆Ω 2v =

2 2πεV ⋅ ∆(h2 ) sin 2 β 0 + ∆(v2 ) cos 2 β 0 sin 2 γ 0 . λ (3.106)

The accuracy of this technique can be estimated in the following manner. If the condition ∆Ωh = ∆Ωv is true in Equation (3.105), we can apply this approximate procedure for the sinc-diagram (see the ﬁrst row in Table 3.1). Curve 2 shown in Figure 3.14 is determined by Equation (3.105) (see the dotted line in Figure 3.14). Curve 2 is very close to the exact determination of the power spectral density (see curve 1 in Figure 3.14) given by Equation (3.95). The case of the Gaussian power spectral density is shown in Figure 3.14 too (curve 3). Comparative analysis made on the basis of Figure 3.14

Copyright 2005 by CRC Press

1598_book.fm Page 119 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Three-Dimensional (Space) Target

119

shows that taking into consideration the real side lobes of the directional diagram is very important during estimation of the power spectral density in the peripheral region. In particular, we cannot approximate the radar antenna directional diagram by the rectangle function, as the rectangle function generates a square waveform power spectral density and cuts off all remainders of the power spectral density of the Doppler ﬂuctuations.

3.3 3.3.1

The Doppler Fluctuations in the Arbitrarily Deflected Radar Antenna General Statements

In this case, it is convenient to go from the coordinates β and γ to the polar coordinate α and θ (see Figure 3.15), in which the position of the radar antenna directional diagram axis is deﬁned by the angles α0 and θ0. If the angle θ0 and the directional diagram width are not so high in value, then sin θ ≈ θ, and we can write ϕ = β − β 0 = θ cos α − β 0 ;

(3.107)

ψ = γ − γ 0 = θ sin α − γ 0 ;

(3.108)

dϕ dψ = θ dα dθ ,

(3.109) ψ

D

A Θ

γ γ0

ϕ θ

θ0 α

0

V

FIGURE 3.15 The coordinate system for variables θ and α.

Copyright 2005 by CRC Press

α0 . β0

β

1598_book.fm Page 120 Monday, October 11, 2004 1:57 PM

120

Signal and Image Processing in Navigational Systems

where β 0 = θ 0 cos α 0

γ 0 = θ 0 sin α 0 .

and

(3.110)

Using Equation (3.78), we can write Ω(ϕ , ψ ) = Ω max cos θ ,

(3.111)

so based on Equation (3.8) we have 2π π

Rg ( τ ) = N

∫ ∫ g (θ cos α − β , θ sin α − γ ) ⋅ e 2

0

0

jΩmax τ cos θ

θ dθ dα .

(3.112)

0 0

Multiplying Equation (3.112) by the factor e jω 0τ and using the Fourier transform, we can write 2π

S g (ω ) ≅

∫ g (θ cos α − β , θ sin α − γ ) dα , 2

0

0

(3.113)

0

where θ = arccos

ω − ω0 ≈ Ω max

2(ω 0 + Ω 0 − ω ) Ω max

(3.114)

and ω ∈[ ω 0 − Ω max , ω 0 + Ω max ] .

(3.115)

Consequently, in the general case, the power spectral density of target return signal Doppler ﬂuctuations is a complex function of the directional diagram and does not coincide with the squared directional diagram because of the high-deﬂected antenna. Reference to Equation (3.113) and Equation (3.114) shows that if the radar antenna is not directional, i.e., if g(ϕ, ψ) 1, then the power spectral density is uniformly distributed within the limits of the interval given by Equation (3.115). This can be explained in the following way. If deﬂection of the radar antenna is not so high, the segment C′ (see Figure 3.11) is directed from segments of straight lines. This difference is greater, the less is the deﬂection of the directional diagram axis from the velocity vector of moving radar. When the segment C′ has an arc shape, the power spectral density at some frequency depends both on ampliﬁcation of the radar antenna along the direction OB and on ampliﬁcation of the radar Copyright 2005 by CRC Press

1598_book.fm Page 121 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Three-Dimensional (Space) Target

121

antenna in other directions. The shape of the power spectral density is changed due to this fact. Consider the case when the directional diagram is symmetrical with respect to its own axis. In this case, the variables ϕ and ψ of the function g(ϕ, ψ) are not separable, as a rule. An exception to this rule is the Gaussian directional diagram. So, in this case, we can write g(ϕ , ψ ) = g( ϕ 2 + ψ 2 ) = g(Θ) ,

(3.116)

where Θ is the angle between the directional diagram axis and the direction to scatterer (see Figure 3.15). The angle Θ is related to the angles θ and θ0 by the following relationships: cos Θ = cos θ 0 cos θ + sin θ 0 sin θ cos(α − α 0 ) .

(3.117)

When the directional diagram width is small, the following relationship: Θ 2 ≅ 2 [1 − cosθ 0 cos θ − sin θ 0 sin θ cos(α − α 0 )]

(3.118)

is true. Moreover, if the angle θ0 is not large, we can write Θ 2 ≅ θ 02 + θ 2 − 2 θ 0 θ cos(α − α 0 ) .

(3.119)

Substituting Equation (3.116)–Equation (3.119) in Equation (3.112) and Equation (3.113), we obtain various formulae for the correlation function and power spectral density, for example 2π

S g (ω ) ≅

2π

∫ g [Θ(α)] dα ≅ ∫ g ( 2

)

θ 02 + θ 2 − 2 θ 0 θ cos(α − α 0 ) dα . (3.120)

2

0

0

Under the conditions β0 = 0

and γ 0 = 0

or

θ0 = 0

and Θ = 0 ,

(3.121)

).

(3.122)

we can write Sg (ω ) ≅ g 2 (θ) = g 2 (arccos

ω −ω o Ωmax

) ≅ g 2(

2⋅

ω 0 + Ω max −ω Ω

The power spectral density is maximal under the condition ω = ω0 + Ωmax and is not symmetric. The effective bandwidth ∆F0.5 of this power spectral

Copyright 2005 by CRC Press

1598_book.fm Page 122 Monday, October 11, 2004 1:57 PM

122

Signal and Image Processing in Navigational Systems

density at the level 0.5 is related to the width of the square of the directional diagram ∆(02.)5 at that level by the following relationship ∆F0.5 =

V ⋅ (∆(02.)5 )2 . 4λ

(3.123)

Because the width ∆(02.)5 is proportional to the wavelength λ, the effective power spectral density bandwidth in Equation (3.123) is also proportional to the wavelength λ, unlike in Equation (3.76). Under the condition θ0 >> ∆a, we can write cos(α – α0) ≈ 1 in Equation (3.120). Then Sg (ω ) ≅ g 2 (θ − θ0 ) = g 2 (

ω 0 + Ωmax cos θ0 − ω Ωmax sin θ0

),

(3.124)

i.e., with the high-deﬂected radar antenna, the power spectral density coincides in shape with the squared radar antenna directional diagram, which is the expected result [compare with Equation (3.74)]. For example, in the case of the circular radar antenna with a uniform distribution of electromagnetic ﬁeld, we can write Sg (ω ) ≅

[

]

2 J1 ( πν ) 4 πν

,

(3.125)

where ν=

3.3.2

ω 0 + Ω max cos θ 0 − ω . Ω max sin θ 0

(3.126)

The Gaussian Directional Diagram

Let the radar antenna directional diagram be Gaussian. Then, from Equation (3.113) it follows that 2π

Sg (ω ) ≅

∫e

p( Ω )cos 2 α + q( Ω )cos( α − α ′ ) − r ( Ω )

dα ,

(3.127)

0

where p(Ω) =

Copyright 2005 by CRC Press

( χ − χ −1 ) Ω ; 2 ∆Ω 0

(3.128)

1598_book.fm Page 123 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Three-Dimensional (Space) Target 2 Ω max Ω ( γ 02 χ 2 + β 02 χ −2 ) ; ∆Ω 02

q(Ω) =

r(Ω) =

123

(3.129)

2 2 −1 ( χ + χ −1 ) Ω ( γ 0 χ + β 0 χ ) Ω max ; + 2 ∆Ω 0 2 ∆Ω 0

(3.130)

γ0 2 ⋅χ ; β0

(3.131)

∆h ; ∆v

(3.132)

tg α ′ =

χ=

Ω = ω 0 + Ω max − ω ; ∆Ω 0 =

(3.133)

Ω max ∆ h ∆ v V∆ h ∆ v . = λ 4π

(3.134)

To determine the power spectral density of the Doppler ﬂuctuations, we can use the series expansion given by Equation (1.15). Then we can write ∞

[

S g (ω ) ≅ I 0 ( p ) ⋅ I 0 ( q ) + 2

∑I

m

]

( p) ⋅ I m (q) cos 2mα ′ ⋅ e − r .

(3.135)

m =1

Equation (3.135) is similar to Equation (1.16) if the following conditions are applied to Equation (1.16) S=

2Ω ∆Ω0

; S0 =

Ωmax ∆Ω0

⋅ θ0 ;

and

ϕ0 = α′ .

(3.136)

Because of this, we can use results discussed in Section 1.2 to determine the power spectral density given by Equation (3.127), as

Sg (ω ) ≅

f

( ). 2Ω ∆Ω0

2Ω ∆Ω0

(3.137)

Using Figure 1.3–Figure 1.8 and Equation (3.137), we can easily construct some curves of the power spectral density for various conditions. In some cases, Equation (3.135) can be fundamentally simpliﬁed. If the directional

Copyright 2005 by CRC Press

1598_book.fm Page 124 Monday, October 11, 2004 1:57 PM

124

Signal and Image Processing in Navigational Systems

diagram is symmetric with respect to its own axis, i.e., if ∆h = ∆v = ∆a , then p(Ω) = 0 and31

(

S g (ω ) ≅ I 0 2 δ 0

Ω ∆Ω0

) ⋅e

Ω − δ2 − ∆Ω 0 0

,

(3.138)

where δ0 =

2 π θ0 . ∆a

(3.139)

Equation (3.138) is similar to Equation (1.21). It is not difﬁcult to prove that under the condition θ0 ≤ 0.4 ∆a, the maximum power spectral density is deﬁned at the frequency ω = ω0 + Ωmax, and their effective bandwidth has the following form:

∆Ω =

V∆(a2) ⋅e λ

2 π θ 02 ∆(a2 )

.

(3.140)

Under the condition θ0 ≥ 0.5 ∆a , we cannot use Equation (3.140), but we use Equation (3.76) and the error is less than 10%. When θ0 = 0, Equation (3.138) can be rewritten in the following form: S g (ω ) ≅ e

Ω − ∆Ω ′0

.

(3.141)

The effective power spectral density bandwidth determined by Equation (3.141) has the following form: ∆Ω = ∆Ω ′0 =

V∆(a2 ) . λ

(3.142)

Usually, the effective bandwidth ∆Ω′0 is very small but not equal to zero, which follows from Equation (3.82). For example, at V = 300 m/sec, ∆a = 2°, and λ = 3 cm, we obtain the effective bandwidth ∆F = 2 Hz. Comparing Equation (3.82) and Equation (3.141), one can see that under the condition θ0 = 90°, the effective power spectral density bandwidth is

2 2π ∆a

times more

than that at θ0 = 0. In other words, the difference is about some hundred times. The normalized power spectral density given by Equation (3.138) is shown in Figure 3.16 at various deﬂections of the directional diagram axis: the angle θ0 is varied from 0 to ∆a. Under the condition θ0 to ∆a , the power spectral Copyright 2005 by CRC Press

1598_book.fm Page 125 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Three-Dimensional (Space) Target

125

density given by Equation (3.80) is shown by the dotted line in Figure 3.16. This power spectral density is correct for high values of the angle θ0 and coincides very well with the middle part of the exactly determined power spectral density. However, there is an essential difference in the remainders. If ∆h ≠ ∆v, but θ0 = 0, then q(Ω) = 0 and we can write Sg (ω ) ≅ I 0

[

( χ − χ −1 ) Ω 2 ∆Ω0

]⋅ e

−

( χ + χ −1 ) Ω 2 ∆Ω0

.

(3.143)

Under the condition ∆h ≈ ∆v , the power spectral density given by Equation (3.143) is not essentially different from that determined by Equation (3.141). The power spectral density given by Equation (3.143) is shown in Figure 3.17. The rate of decrease of the power spectral density Sg(ω) depends mainly on the greatest values of ∆h and ∆v if the difference between ∆h and ∆v is high. The effective power spectral density bandwidth given by Equation (3.143) then takes the following form: ∆Ω =

V∆ h ∆ v . λ

(3.144)

Sg (ω)

1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2

5

6

4 3 ω − ω0 − Ωmax

2

0.1

1 −20 −18 −16 −14 −12 −10 −8.0 −6.0 −4.0 −2.0

∆Ω0 0

FIGURE 3.16 The power spectral density of the Doppler ﬂuctuations. The Gaussian directional diagram is deﬂected, ∆h = ∆v = ∆a: (1)

Copyright 2005 by CRC Press

θ0 ∆a

= 0 ; (2)

θ0 ∆a

= 0.2 ; (3)

θ0 ∆a

= 0.4 ; (4)

θ0 ∆a

= 0.6 ; (5)

θ0 ∆a

= 0.8 ; (6)

θ0 ∆a

=1.

1598_book.fm Page 126 Monday, October 11, 2004 1:57 PM

126

Signal and Image Processing in Navigational Systems Sg (ω) 1.0 0.9 0.8 0.7 0.6 0.5 1

0.4

2 0.3

3

0.2 ω − ω0 − Ωmax

0.1

∆ Ω0

−4.0 −3.5 −3.0 −2.5 −2.0 −1.5 −1.0 −0.5

0

FIGURE 3.17 The power spectral density of the Doppler ﬂuctuations. The Gaussian directional diagram of radar antenna is not deﬂected, ∆h ≠ ∆v: (1)

∆h ∆v

= 1 ; (2)

∆h ∆v

= 2 ; (3)

∆h ∆v

=5.

Under the condition ∆h = ∆v = ∆a , the effective bandwidth ∆Ω determined by Equation (3.144) coincides with the effective power spectral density bandwidth determined by Equation (3.141). In the case of the Gaussian radar antenna directional diagram and under the conditions β0 = 0 and γ0 = 0, it is not difﬁcult to deﬁne the correlation function of the Doppler ﬂuctuations using Equation (3.112).

3.3.3

Determination of the Power Spectral Density

As was discussed in Section 3.1.3, determination of the power spectral density of the target return signal Doppler ﬂuctuations reduces to determination of the total target return signal power from scatterers giving the same shift in frequency. This technique is not general, but it is very clear from the physical viewpoint. Using this technique in many practical cases, we can determine the power spectral density without deﬁning the correlation function of the Doppler ﬂuctuations.25,45 For example, in the considered problem, an annular domain (see Figure 3.10) is the geometrical center of scatterers

Copyright 2005 by CRC Press

1598_book.fm Page 127 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Three-Dimensional (Space) Target

127

with the same Doppler shift in frequency. The target return signal power from the scatterer scanned under the angle Θ (see Figure 3.15) is proportional to the squared radar antenna directional diagram g2(Θ). The total target return signal power from the annular domain (θ, θ + dθ; ρ, ρ + 0.5 cτp) has the following form:

∫

dp = m0 p1 g 2 (Θ) dQ ,

(3.145)

where m0 is the average number of scatterers per volume unit; p1 is the target return signal power from the individual scatterer with effective scattering area S1 when the scatterer is placed on the directional diagram axis [see Equation (2.107)]; Q is the integrated domain [see Equation (2.108)]; dQ = ρ 2 sin θ dρ dθ dα

(3.146)

is the volume element. Under the conditions ρ = const and θ = const, which are satisﬁed within the considered volume, we can write 2π

∫

dp = m0 p1 (0.5c τ p )ρ2 sin θ dθ g 2 [Θ(α )] dα .

(3.147)

0

Equation (3.147) together with Equation (3.78) is the parametric form of the power spectral density; the angle θ is the parameter. Reference to Equation (3.78) shows that dΩ = − Ω max sin θ dθ .

(3.148)

Going from the power dp to the power spectral density S(Ω) =

dp [Θ(Ω)] , dΩ

(3.149)

we obtain the well-known formula [see Equation (3.120)]. This technique can be successfully used in the determination of the power spectral density of the Doppler ﬂuctuations under scanning of the two-dimensional (surface) target (see Section 4.8) and in the study of some forms of chaotic motion of scatterers (see Section 7.2).

Copyright 2005 by CRC Press

1598_book.fm Page 128 Monday, October 11, 2004 1:57 PM

128

Signal and Image Processing in Navigational Systems

3.4

The Total Power Spectral Density with the Pulsed Searching Signal

3.4.1

General Statements

Reference to Equation (3.12) shows that if the three-dimensional (space) target is scanned by the pulsed searching signal of moving radar, the total normalized correlation function R(τ) of target return signal ﬂuctuations is deﬁned by the product of the periodic normalized correlation function Rp(τ) of the ﬂuctuations in the radar range, or in the distance between the radar and scatterer, and the nonperiodic normalized correlation function Rg(τ) of the Doppler ﬂuctuations (see Figure 3.1–Figure 3.3). Naturally, in this case, the total normalized correlation function R(τ) is not a periodic function because with an increase in the value of τ = nTp , the waves of the normalized correlation function R(τ) decrease in value due to the Doppler ﬂuctuations. This destruction of correlation is strong; the value of n is high (see Figure 3.3a). The total power spectral density S(ω) is deﬁned by convolution of the linear power spectral density Sp(ω) given by Equation (3.15) and the continuous power spectral density Sg(ω) given by Equation (3.43) (see Figure 3.1–Figure 3.3): ∞

S(ω ) ≅

∫ S (x) ∗ S (ω − x) dx =∑ ∫ S p

en p

g

( x) ⋅ δ( x − nΩ′p ) ⋅ Sg (ω − x) dx

n= 0

∞

=

∑ n= 0

∞

∑ S (ω − nΩ′ ) .

Spen (nΩ′p ) ⋅ Sg (ω − nΩ′p ) ≈ Spen (ω )

g

p

(3.150)

n= 0

We have to use the result of convolution at the frequency ω0 + Ω0 [see Equation (3.150)] if this fact has not been taken into consideration in the power spectral densities Sp(ω) or Sg(ω) . As a rule, the power spectral density Sg(ω – nΩp′ ) is very narrow in comparison with the power spectral density Spen (ω ) . The total power spectral density S(ω) can be approximately considered as the product of the wedge-like Doppler power spectral densities

∑

∞ n= 0

Sg (ω − nΩ′p ) and the envelope Spen (ω ) of the power spectral density

Sp(ω) of the ﬂuctuations in the radar range (see Figure 3.3b). These power spectral densities have been determined in Section 3.1–Section 3.3 for various cases. Deﬁnition of the total power spectral density based on Equation (3.150) is not difﬁcult. For example, with the square waveform pulsed searching signal and the Gaussian radar antenna directional diagram, the deﬂection

Copyright 2005 by CRC Press

1598_book.fm Page 129 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Three-Dimensional (Space) Target

129

of which is very high, reference to Equation (3.19) and Equation (3.80) shows that the total power spectral density of the target return signal ﬂuctuations can be written in the following form: ∞

S(ω ) ≅

∑

sinc 2 (nΩ′p , 0.5τ p ) ⋅ e

−

π ∆Ω2

⋅ ( ω − ω 0 − Ω0 − nΩ′p )2

,

(3.151)

n= 0

where ∆Ω is determined by Equation (3.81).

3.4.2

Interperiod Fluctuations in the Glancing Radar Range

Consider the total normalized correlation function R(τ) given by Equation (3.12). Reasoning that τ = nTp′ , we can obtain the normalized correlation function of the interperiod target return signal ﬂuctuations in the pure form based on the total normalized correlation function R(τ). Under the condition τ = nTp′ , we can write Rp(τ) = 1 and R(τ) = Rg(τ). The condition τ = nTp′ means that the correlation function is determined for widely spaced instants of time on time periods that are compressed n times or expanded µ = 1 + c r times of periods.46 This correlation function characterizes the Doppler ﬂuctuations caused by scanning the same pulse volume with moving radar. In other words, this correlation function takes into consideration changes in distance between the radar and scatterers during the time nTp′ (see Figure 3.18). The power spectral density of these target return signal ﬂuctuations coincides with the power spectral density Sg(ω) shifted in frequency by ω0 + Ω0. The power spectral density Sg(ω) is investigated in Section 3.2 and Section 3.3. In other words, we can state that the power spectral density of these ﬂuctuations coincides with the main wave of the total power spectral density S(ω). 2V

3.4.3

Interperiod Fluctuations in the Fixed Radar Range

It is worthwhile to consider the interperiod ﬂuctuations not only in the glancing radar range, but also in the ﬁxed radar range, i.e., under the condition τ = nTp as well.47,48 This means that the correlation function of the ﬂuctuations is determined for widely spaced instants of time on n undistorted periods, i.e., for the instants of time ﬁxed with respect to the instant of time of generation of the pulsed searching signal. In this case, the correlation function characterizes the ﬂuctuations arising by scanning the pulse volume, which is located at a ﬁxed distance from the moving radar. In other words, we can state that the pulse volume moves with the radar (or aircraft). Unlike the previous case in Section 3.4.2, the ﬂuctuations in the ﬁxed radar range are caused by two reasons. The ﬂuctuations caused by the moving pulse volume (the distance between the moving radar and the moving pulse volume

Copyright 2005 by CRC Press

1598_book.fm Page 130 Monday, October 11, 2004 1:57 PM

130

Signal and Image Processing in Navigational Systems c τp 2 ρ

ρ − Vr nT ′p 0

FIGURE 3.18 Fluctuations forming in the glancing radar range.

ρ

ρ

FIGURE 3.19 Fluctuations forming in the ﬁxed radar range.

is ﬁxed) arise in addition to the Doppler ﬂuctuations. These ﬂuctuations are, as a rule, identical to the ﬂuctuations caused by exchange of scatterers in the case of square waveform pulsed searching signals. In the case of the pulsed searching signal with an arbitrary shape, the boundaries of the pulse volume are not clear. This can be explained by amplitude changes in the elementary signals due to their modulation by the envelope of the pulsed searching signal amplitude with moving radar. It is necessary to consider the interperiod ﬂuctuations both in the ﬁxed radar range and in the glancing radar range because in interperiod signal processing by navigational systems, the target return signal at the receiver

Copyright 2005 by CRC Press

1598_book.fm Page 131 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Three-Dimensional (Space) Target

131

or detector input can be shifted in time both by the value τ = nTp (radar motion is not taken into consideration) and by the value τ = nTp′ (radar motion is taken into consideration). Moreover, in some cases arising in practice, for example, when tracking a point target against a background of the three-dimensional (space) or the two-dimensional (surface) target, signal processing must be carried out at intervals coinciding with the value of Tp′′ , not of Tp or T′p′ where the time interval Tp′′ takes into consideration the moving tracking target. We do not consider this case. It is a continuation of the next case. Replace the radial velocity Vr0 with the approach velocity Vr0 + Vrt of the radar and target. Assume that τ = nTp. Reference to Equation (3.5)–Equation (3.8) shows that R ( τ) = Rpen ( µτ) ⋅ Rg ( τ) ,

(3.152)

where µ = µ −1=

2 Vr0 c

=

2V ⋅ cos β 0 cos γ 0 << 1 ; c

∫ P ( z) ⋅ P ( z + ( µτ) = ∫ P (z) dz

2 Vr

∗

en p

R

0

c

(3.153)

) dz .

2

(3.154)

Thus, the normalized correlation function of the interperiod ﬂuctuations in the ﬁxed radar range is deﬁned by the product of the normalized correlation function Rg(τ) of the interperiod ﬂuctuations in the glancing radar range and the normalized correlation function Rpen ( µτ) of the interperiod ﬂuctuations with the highly expanded scale — µτ instead of µτ. The normalized correlation function Rpen ( µτ) deﬁnes the ﬂuctuations caused by exchange of scatterers. The correlation interval of the target return signal ﬂuctuations caused by exchange of scatterers is equal to the time required for the pulse volume — the resolution element — moving with velocity Vr0 to be renewed, and the power spectral density of the ﬂuctuations caused by exchange of scatterers depends on the pulsed searching signal shape and is determined by Equation (3.17), in which µ is replaced with µ :

Sp (ω ) ≅

Copyright 2005 by CRC Press

∫

P(t) ⋅ e

−

jω t µ

2

dt

.

(3.155)

1598_book.fm Page 132 Monday, October 11, 2004 1:57 PM

132

Signal and Image Processing in Navigational Systems

For example, with the square waveform pulsed searching signal, the normalized correlation function given by Equation (3.154) takes the following form Rpen ( µτ) = 1 −

|µτ| | τ| = 1− τp τc

| τ| ≤ τ c ,

at

(3.156)

where τc =

τp µ

=

cτ 2 V cos β 0 cos γ 0

>> τ p

(3.157)

is the time required for the radar moving with velocity V to traverse a distance

c τp 2 cos β 0 cos γ 0

.

The corresponding power spectral density of the ﬂuctuations coincides with the power spectral density of the square waveform pulsed searching signal with length τc given by Equation (3.157): Sp (ω ) ≅ sinc 2 (0.5ωτ c ) .

(3.158)

The effective bandwidth ∆F = τ1c of the power spectral density of the target return signal ﬂuctuations is independent both of the wavelength and the radar antenna directional diagram width and is deﬁned by the pulsed searching signal duration, velocity of moving radar, and directional diagram orientation. When the directional diagram is not deﬂected, the effective bandwidth ∆F of the power spectral density is maximal, but the maximum is very steep. Consider this example: at V = 300 m/sec, τp = 0.5 µsec, β0 = 0°, and γ0 = 0,we obtain the effective bandwidth ∆F = 4 Hz; if β0 = 45° and γ0 = 45° we obtain the effective bandwidth ∆F = 2 Hz. Usually, the ﬂuctuations caused by exchange of scatterers are very slow. In the majority of cases, we can neglect these ﬂuctuations in comparison with the Doppler ﬂuctuations, which have a power spectral density bandwidth that is 10 or 102 times more [see the example in Section 3.2.2, Equation (3.85)]. However, there are exceptions to this rule; for example, if the pulsed searching signal duration is very low in value, or in the case of the frequency-modulated pulsed signal, when the power spectral density Sp(ω) in the radar range is expanded, or if the directional diagram is not deﬂected and the power spectral density Sg(ω) is narrowed down [see the example in Section 3.3.2, Equation (3.141)]. In the case of the square waveform linear-frequency modulated pulsed signals with the deviation ∆ωM , we can write based on Equation (3.31)

Copyright 2005 by CRC Press

1598_book.fm Page 133 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Three-Dimensional (Space) Target

R ( µτ) = en p

| sin[0.5∆ω M µτ(1 − |µτ τ p )]

0.5∆ω M µτ

.

133

(3.159)

Reference to Equation (3.160) shows that as ∆ωM → 0, we obtain the normalized correlation function given by Equation (3.156) and, at ∆ω M >> 2τ pπ , the normalized correlation function is determined by the function sinc(0.5∆ω M µτ), with the correlation interval given by τc =

cτp c 1 , = = µ ∆ f M 2 Vr0 ∆ f M 2 Vr0 D

(3.160)

where D = ∆fM τp. Therefore, the correlation interval given by Equation (3.160) is D times less than the correlation interval given by Equation (3.157). Consequently, the effective bandwidth is D times greater than the effective power spectral density bandwidth given by Equation (3.158). The shape of the power spectral density tends to approach the square waveform shape. For an arbitrary value of D, the power spectral density coincides with the envelope of the power spectral density given by Equation (3.32) and the correlation interval is determined by Equation (3.36) if τp is replaced with

τp µ

. Under these con-

ditions, we can use Figure 3.5–Figure 3.7. In the cases of the Gaussian pulsed searching signal without linear-frequency modulation and with linear-frequency modulation, both the normalized correlation function and the power spectral density of the target return signal ﬂuctuations are deﬁned by the Gaussian law. The effective bandwidth and correlation interval of the power spectral density have the following form:

∆F =

1 = τc

2 (1 + D2 ) Vcosβ 0 cos γ 0 . cτp

(3.161)

When D = 0, the effective bandwidth ∆F is 2 times less than the effective power spectral density bandwidth given by Equation (3.158). If D >> 1, the effective power spectral density bandwidth increases D times. If the pulsed searching signal and directional diagram are Gaussian, then in the case of high-deﬂected radar antenna, where the power spectral density of the Doppler ﬂuctuations is determined by Equation (3.80), the total power spectral density of the slow ﬂuctuations given by Equation (3.152) is Gaussian too, with the effective bandwidth given by ∆F = ∆F12 + ∆F22 ,

Copyright 2005 by CRC Press

(3.162)

1598_book.fm Page 134 Monday, October 11, 2004 1:57 PM

134

Signal and Image Processing in Navigational Systems

where the effective bandwidth ∆F1 is determined by Equation (3.81) and the effective bandwidth ∆F2 is given by Equation (3.161). Equation (3.162) allows us to estimate an extension of the power spectral density of the ﬂuctuations caused by exchange of scatterers without any difﬁculty. When the radar antenna is not deﬂected, the power spectral density of the Doppler ﬂuctuations has the form of the exponential function given by Equation (3.141) and the convolution of this power spectral density and the Gaussian power spectral density of the ﬂuctuations caused by exchange of scatterers has the following form

{

S(ω ) ≅ 1 + Φ  π 

(

Ω ∆Ω

−

∆Ω2 2 π∆Ω′0

)} ⋅ e

− Ω

∆Ω0′

,

(3.163)

where Ω = ω 0 + Ω max − ω ,

(3.164)

∆Ω2 is determined by Equation (3.161), and ∆Ω0′ is given by Equation (3.141). The power spectral density given by Equation (3.163) is shown in Figure 3.20 under the condition ∆Ω2 = ∆Ω0′ . S (ω)

1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 ω − ω0 − Ωmax

0.1

∆ Ω0 −4.0 −3.5 −3.0 −2.5 −2.0 −1.5 −1.0 −0.5

0

0.5

1.0

FIGURE 3.20 The total power spectral density. The Gaussian directional diagram, θ0 = 0.

Copyright 2005 by CRC Press

1598_book.fm Page 135 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Three-Dimensional (Space) Target 3.4.4

135

Irregularly Moving Radar

Consider the case when the radar moves varying both in the value and direction of velocity. Because the Doppler effect is directly initiated by the moving radar, changes in the velocity of moving radar result in changes in all consequences of the effect. With the simple harmonic searching signal, the function V(t) shows that the average Doppler frequency given by Equation (3.75) and the effective power spectral density bandwidth of the Doppler ﬂuctuations given by Equation (3.76) depend on time. The process becomes nonstationary. With the pulsed searching signal, in addition to this, the τ

T

values of the pulse duration τ ′p = µp and the pulse period Tp′ = µp , where µ is determined by Equation (3.11), become functions of time. The ﬁrst phenomenon is inﬁnitesimal, as a rule, but the second phenomenon appears in the investigation of the interperiod ﬂuctuations with ﬁxed radar range as a function of time for the effective power spectral density bandwidth of the target return signal ﬂuctuations caused by exchange of scatterers [see Equation (3.157)]. Changes in the velocity along a direction of moving radar lead us to amplitude modulation of elementary signals, which is similar to amplitude modulation caused by radar antenna scanning. Suppose the radar moves in a horizontal plane with constant velocity V along an arc with radius ρ. As a result, the angle shifts of scatterers with respect to the horizontal-coverage directional diagram ∆ϕ = Vρτ , where Vρ is the angular velocity of the directional diagram, arise. In this case, the formulae in Equation (3.5) and Equation (3.12) are true, but instead of Equation (3.8) we have to use the general formula in Equation (2.94). The general formula in Equation (2.94) takes into consideration the angle shifts of scatterers. The study procedure is the same as in the case of simultaneous radar movement and radar antenna scanning. However, there is a distinction in principle. Because the position of the directional diagram with respect to the velocity vector is not variable for the considered case, the effective power spectral density bandwidth of the Doppler ﬂuctuations is independent of the parameter t (time), and if the condition ρ = const is satisﬁed, the process is stationary. With the deﬂected Gaussian directional diagram, the power spectral density of the target return signal ﬂuctuations is Gaussian too and the effective power spectral density bandwidth has the following form ∆F = ∆FD2 + ∆Fρ2 ,

(3.165)

where ∆FD is the effective power spectral density bandwidth of the Doppler ﬂuctuations given by Equation (3.82); ∆Fρ =

Copyright 2005 by CRC Press

V cos γ 0 2 ρ ∆h

(3.166)

1598_book.fm Page 136 Monday, October 11, 2004 1:57 PM

136

Signal and Image Processing in Navigational Systems

is the effective power spectral density bandwidth of the target return signal ﬂuctuations caused by curvature of the trajectory of aircraft, for example. As a rule, the value of ∆Fρ is very low.

3.5

Conclusions

We summarize brieﬂy the main results discussed in this chapter. In the study of the target return signal ﬂuctuations caused by the moving radar, the total correlation function is a nonstationary separable process, i.e., the normalized correlation function is independent of the time parameter t. All information regarding a nonstationary state is included in the power of the target return signal given by Equation (2.78). With the continuous nonmodulated pulsed searching signal, the normalized correlation function Rp(τ) 1 and the total normalized correlation function of target return signal ﬂuctuations coincides with the normalized correlation function of the slow ﬂuctuations. The rapid ﬂuctuations are absent. If the radar is stationary, the condition Rg(τ) 1 is true and there are only the rapid (intraperiod) ﬂuctuations under the condition µ = 1. The slow ﬂuctuations are absent. Radar motion deﬁned by changeover from the argument τ to the argument µτ in the normalized correlation function Rp(τ) of the ﬂuctuations given by Equation (3.6) implies a compression (µ < 1) or expansion (µ > 1) of the time scale µ times. This is a natural manifestation of the Doppler effect, which is accompanied by changes in the pulsed searching signal duration and iteration frequency, in addition to changes in the carrier frequency under radar antenna scanning. The shape of the normalized correlation function and the effective power spectral density bandwidth of intraperiod ﬂuctuations are completely deﬁned by the shape and duration of the pulsed searching signal. The Doppler ﬂuctuations caused by the moving radar are deﬁned by the envelope of the normalized correlation function of the ﬂuctuations in the radar range given by Equation (3.12). The normalized correlation function Rg(τ) characterizes the interperiod ﬂuctuations. The power spectral density at the relative Doppler frequency is formed by summing the powers of the target return signals from those scatterers that are placed on the surface of the cone deﬁned near the velocity vector of moving radar and with apex angle 2θ. When the radar antenna axis is highly deﬂected from the direction of moving radar and the variables ϕ′ and ψ′ are separable in the radar antenna directional diagram, then the power spectral density of the Doppler ﬂuctuations coincides in shape with the square of the directional diagram in the plane crossing a direction of moving radar and the directional diagram axis. The effective power spectral density bandwidth is deﬁned by the squared directional diagram width by power in the plane of radar antenna deﬂection. If the radar antenna axis is deﬂected in an arbitrary way from the direction Copyright 2005 by CRC Press

1598_book.fm Page 137 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Three-Dimensional (Space) Target

137

of moving radar, the power spectral density is a complex function of the directional diagram and does not coincide with the squared directional diagram. If the three-dimensional (space) target is scanned by pulsed searching signals of moving radar, the total normalized correlation function R(τ) is deﬁned by the product of the periodic normalized correlation function Rp(τ) of the ﬂuctuations in the radar range and the nonperiodic normalized correlation function Rg(τ) of the Doppler ﬂuctuations. In this case, the total normalized correlation function of the target return signal ﬂuctuations is not a periodic function. The total power spectral density is deﬁned by convolution of the linear power spectral density of the ﬂuctuations in the radar range and the continuous power spectral density of the Doppler ﬂuctuations. The normalized correlation function of the interperiod ﬂuctuations in the ﬁxed radar range is deﬁned by the product of the normalized correlation function Rg(τ) of the interperiod ﬂuctuations in the glancing radar range and the normalized correlation function Rpen ( µτ) of the interperiod ﬂuctuations.

References 1. Ward, K., Baker, C., and Watts, S., Maritime surveillance radar. Part 1: radar scattering from the ocean surface, in Proc. Inst. Elect. Eng. F, Vol. 137, No. 2, 1990, pp. 51–62. 2. Hall, H., A new model for impulsive phenomena: application to atmosphericnoise communication channel, Technical Report 3412-8, Stanford University, Stanford, CA, 1966. 3. Di Bisceglie, M. and Galdi, C., Random walk based characterization of radar backscatterer from the sea surface, in Radar, Sonar, and Navigation, IEEE Proceedings, Vol. 145, No. 4, 1993, pp. 216–225. 4. Levanon, N., Radar Principles, Wiley, New York, 1988. 5. Pentini, A., Farina, A., and Zirilli, F., Radar detection of targets located in a coherent K-distributed clutter background, in Proceedings IEEE, Vol. 139, June 1992, pp. 341–358. 6. Rihaczek, A., Principles of High-Resolution Radar, Peninsala, San Jose, CA, 1985. 7. Doisy, Y., Derauz, L., Beerens, P., and Been, R., Target Doppler estimation using wideband frequency modulated signals, IEEE Trans., Vol. SP-48, No. 5, 2000, pp. 1213–1224. 8. Kramer, S., Doppler and acceleration tolerances of high-gain, wideband linear FM correlation sonars, in Proceedings of the IEEE, Vol. 55, No. 5, 1967, pp. 627–636. 9. Farina, A., Antenna-Based Signal Processing Techniques for Radar Systems, Artech House, Norwood, MA, 1992. 10. Baculev, P. and Slepin, V., Methods and Apparatus of Selection for Moving Targets, Soviet Radio, Moscow, 1986 (in Russian). 11. Parsons, J., The Mobile Radio Propagation Channel, John Wiley & Sons, New York, 1996.

Copyright 2005 by CRC Press

1598_book.fm Page 138 Monday, October 11, 2004 1:57 PM

138

Signal and Image Processing in Navigational Systems

12. Ali, I., Al-Dhair, N., and Hershey, J., Doppler characterization for LEO satellites, IEEE Trans., Vol. COM-46, No. 3, 1998, pp. 309–313. 13. Ward, J., Space-time adaptive processing for airborne radar, Lincoln Labs Technical Report 1015, MIT, Cambridge, MA, 1994. 14. Collins, T. and Atteins, P., Doppler-sensitive active sonar pulse designs for reverberation processing, in Radar, Sonar, and Navigation, IEEE Proceedings, Vol. 145, No. 12, 1998, pp. 1215–1225. 15. Bretthorst, G., Bayesian Spectrum Analysis and Parameter Estimation, SpringerVerlag, New York, 1988. 16. Costas, J., A study of a class of detection waveforms having nearly ideal rangeDoppler ambiguity properties, in Proceedings of the IEEE, Vol. 72, No. 6, 1984, pp. 996–1009. 17. Muirhead, R., Aspects of Multivariate Statistical Theory, John Wiley & Sons, New York, 1982. 18. Haykin, S., Non-Linear Methods of Spectral Analysis, Springer-Verlag, New York, 1979. 19. Kay, S., Modern Spectral Estimation: Theory and Application, Prentice Hall, Englewood Cliffs, NJ, 1988. 20. Rozenbach, K. and Ziegenbein, J., About the effective Doppler sensitivity of certain non-linear chirp signals (NLFM), in Proceedings of the Low Frequency Active Sonar Conference, La Spezia, Italy, May 24–28, 1993, pp. 571–579. 21. Shanmugan, K. and Breipohl, A., Random Signals: Detection, Estimation, and Data Analysis, John Wiley & Sons, New York, 1988. 22. Pillai, S., Array Signal Processing, Springer-Verlag, New York, 1998. 23. Pahlavan, K. and Levesque, A., Wireless Information Networks, John Wiley & Sons, New York, 1995. 24. Haykin, S., Adaptive Filter Theory, 3rd ed., Prentice Hall, Englewood Cliffs, NJ, 1996. 25. Feldman, Yu, Determination of spectrum of target return signals, Problems in Radio Electronics, Vol. OT, No 6, 1959, pp. 22–38 (in Russian). 26. Stoica, P. and Moses, R., Introduction to Spectral Analysis, Prentice-Hall, Englewood Cliffs, NJ, 1997. 27. Proakis, J. and Manolakis, D., Digital Signal Processing Principles, Algorithms, and Applications, Prentice Hall, Englewood Cliffs, NJ, 1995. 28. Brillinger, D., Time Series: Data Analysis and Theory, Holden-Day, San Francisco, CA, 1981. 29. Johnson, N. and Kotz, S., Distributions in Statistics: Continuous Univariate Distributions, Vol. 2, John Wiley & Sons, New York, 1970. 30. Papoulis, A., Probability, Random Variables, and Stochastic Processes, McGrawHill, New York, 1984. 31. Feldman, Yu, Gidaspov, Yu, and Gomzin, V., Moving Target Tracking, Soviet Radio, Moscow, 1978 (in Russian). 32. Armond, N., Correlation function of waves scattered by rough surfaces, Radio Eng. Electron. Phys., Vol. 30, No. 7, 1985, pp. 1307–1311 (in Russian). 33. Watts, Ed., Radar clutter and multipath propagation, in Proceedings of IEEF, Vol. 138, April 1994, pp. 187–199. 34. Ward, K., Compound representation of high resolution sea clutter, Electron. Lett., Vol. 17, No. 16, 1981, pp. 561–563. 35. Poor, V., An Introduction to Signal Detection and Estimation, Springer-Verlag, New York, 1988.

Copyright 2005 by CRC Press

1598_book.fm Page 139 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Three-Dimensional (Space) Target

139

36. Carter, G., Coherence and Time Delay Estimation, IEEE Press, New York, 1993. 37. Tsao, J. and Steinberg, D., Reduction of side-lobe and speckle artifacts in microwave imaging: the CLEAN technique, IEEE Trans., Vol. AP-36, No. 2, 1988, pp. 543–556. 38. Kalson, S., An adaptive array detector with mismatched signal detection, IEEE Trans., Vol. AES-28, No. 1, 1992, pp. 195–207. 39. Rappaport, T., Wireless Communications Principles and Practice, Prentice Hall, Upper Saddle River, NJ, 1996. 40. Jourdain, G. and Henrioux, J., Use of large bandwidth-duration binary phase shift keying signals in target delay Doppler measurements, J. Acoust. Soc. Amer., Vol. 90, No. 1, 1991, pp. 299–309. 41. Stark, H. and Woods, J., Probability, Random Processes, and Estimation Theory for Engineers, Prentice Hall, Englewood Cliffs, NJ, 1986. 42. Porat, B., Digital Processing of Random Signals, 5th ed., Prentice Hall, Englewood Cliffs, NJ, 1994. 43. Gotwols, B., Chapman, R., and Sterner II, R., Ocean radar backscatterer statistics and the generalized log normal distribution, in Proceedings of PIERS94, The Netherlands, July 11–15, 1994, pp. 1028–1031. 44. Feldman, Yu, Nonlinear transformations of Doppler spectra, Problems in Radio Electronics, Vol. OT, No. 5, 1980, pp. 3–14 (in Russian). 45. Borkus, M., Energy spectrum of target return signal from atmosphere aerosol scatterers, Problems in Radio Electronics, Vol. OT, No. 1, 1977, pp. 43–50 (in Russian). 46. Kroszczinski, J., Pulse compression by means of linear-period modulation, in Proceedings of the IEEE, Vol. 57, No. 7, 1969, pp. 1260–1266. 47. Tseng, C. and Gifﬁths, L., A uniﬁed approach to the design of linear constraints in minimum variance adaptive beamformers, IEEE Trans., Vol. AP-40, No. 6, 1992, pp. 1533–1542. 48. Wax, M. and Anu, Y., Performance analysis of the minimum variance beamformer, IEEE Trans., Vol. SP-44, No. 4, 1996, pp. 928–937.

Copyright 2005 by CRC Press

1598_book.fm Page 141 Monday, October 11, 2004 1:57 PM

4 Fluctuations under Scanning of the Two-Dimensional (Surface) Target by the Moving Radar

4.1

General Statements

As explained in Chapter 3, we assume that the radar antenna is stationary, ∆ϕsc = ∆ψsc = 0, and we consider the long-range area of the radar antenna directional diagram ∆ϕrm = ∆ψrm = 0. We assume that the radar moves rectilinearly and uniformly with the velocity V, i.e., ∆ = –V · τ. Based on Equations (2.122), (2.127), (2.128), and (2.165), we can write ∞

R en (t , τ) = p0

∑ ∫∫ P[t −

2 ρ( ψ ) c

− 0.5(µτ − nTp )] ⋅ P ∗[t −

2 ρ( ψ ) c

+ 0.5(µτ − nTp )]

n= 0

× g 2 (ϕ , ψ ) ⋅ S°(β 0 , ψ + γ 0 )sin(ψ + γ 0 ) ⋅ e jΩ(ϕ ,ψ ) τ dϕ dψ,

(4.1)

where Ω(ϕ , ψ ) = Ω max [cos ε 0 cos( β 0 +

ϕ cos γ

) cos(ψ + γ 0 ) − sin ε 0 sin(ψ + γ 0 )] ;(4.2)

µ = 1 + 2Vr c −1 = 1 + 2Vc −1 (cos ε 0 cos β 0 cos γ − sin ε 0 sin γ ) ;

(4.3)

γ- = γ0 with the continuous searching signal; γ- = γ* with the pulsed searching signal; the power p0 of the target return signal is determined by Equation (2.122); and the maximum Doppler frequency Ωmax is given by Equation (3.10). Using Equation (4.1), we can investigate the case of both the searching simple harmonic signal and the pulsed searching signal.1,2 At the pulsed

141 Copyright 2005 by CRC Press

1598_book.fm Page 142 Monday, October 11, 2004 1:57 PM

142

Signal and Image Processing in Navigational Systems

searching signal and conditions S°(β0, ψ + γ0) ≈ S°(β0, γ*) and sin(ψ + γ0) ≈ sin γ*, based on Equation (4.1), we can write R en (t , τ) = p(t) ⋅ R en (t , τ) ,

(4.4)

where ∞

R en (t , τ) = N

∑ ∫∫ P[t −

2 ρ( ψ ) c

− 0.5(µτ − nTp )] (4.5)

n= 0

× P ∗[t −

2 ρ( ψ ) c

+ 0.5(µτ − nTp )]g 2 (ϕ , ψ ) ⋅ e jΩ(ϕ ,ψ ) τ dϕ dψ

The power p(t) of the target return signal was studied in more detail in Section 2.5.3. We assum that, with the pulsed searching signal, the conditions given by Equations (2.129) and (2.130) are satisﬁed and we can write R en (t , τ) = Rβ (t , τ) ⋅ Rγ (t , τ)

(4.6)

instead of Equation (4.5), where

∫

Rβ (t , τ) = N g h2 (ϕ , ψ ∗ ) ⋅ e

jΩβ ( ϕ ) τ

dϕ ;

(4.7)

∞

Rγ (t , τ) = N

∑ ∫ P[t −

2 ρ( ψ ) c

− 0.5(µτ − nTp )] ⋅ P ∗[t −

2 ρ( ψ ) c

+ 0.5(µτ − nTp )]

n= 0

× gv2 (ψ ) ⋅ e

− j ( ψ − ψ ∗ )Ω γ τ

dψ (4.8)

Ωβ (ϕ) = Ω max [cos ε 0 cos( β 0 +

ϕ cos γ ∗

) cos γ ∗ − sin ε 0 sin γ ∗ ] ;

(4.9)

Ω γ = Ω max (cos ε 0 cos β 0 sin γ ∗ − sin ε 0 cos γ ∗ ) ;

(4.10)

Ω(ϕ , ψ ) ≈ Ωβ (ϕ) − ( ψ − ψ ∗ )Ω γ .

(4.11)

Compare with Equation (4.2). The azimuth-normalized correlation function Rβ(t, τ) given by Equation (4.7) takes into consideration the slow target return signal ﬂuctuations caused by differences in the Doppler frequencies in the azimuth plane. The aspectangle normalized correlation function Rγ(t, τ) given by Equation (4.8) takes Copyright 2005 by CRC Press

1598_book.fm Page 143 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

143

into consideration the slow target return signal ﬂuctuations, which in turn are caused by differences in the Doppler frequencies in the aspect-angle plane, and the rapid target return signal ﬂuctuations, which are caused by the propagation of the pulsed searching signal along the scanned surface of the two-dimensional target.3–5

4.2

The Continuous Searching Nonmodulated Signal

Assuming that the condition given by Equation (2.109) is satisﬁed in Equation (4.1), and omitting the summation sign, we can write the correlation function in the following form6,7 R en (t , τ) = p(t) ⋅ Rgen ( τ) ,

(4.12)

where the power p(t) of the target return signal is determined by Equation (2.166) and Rgen (τ) = N

∫∫ g (ϕ, ψ) ⋅ S°(β , ψ + γ )sin(ψ + γ ) ⋅ e 2

0

0

jΩ( ϕ , ψ ) τ

0

dϕ dψ . (4.13)

Comparing Equation (4.13) with Equation (3.8), one can see that Equation (4.13) follows from Equation (3.8) if we replace the function g2(ϕ, ψ) in Equation (3.8) with the function ~g2(ϕ, ψ) that can be determined in the following form g 2 (ϕ , ψ ) = g 2 (ϕ , ψ ) ⋅ S°(β 0 , ψ + γ 0 )sin(ψ + γ 0 )

(4.14)

and assume that ε0 = 0 in Equation (4.2). The function g~(ϕ, ψ) can be considered as the generalized radar antenna directional diagram.8,9 For this reason, certain results obtained in Chapter 3 can be used in the investigations in this chapter, replacing the function g(ϕ, ψ) with the generalized function g~(ϕ, ψ). We consider that the vector of velocity of the moving radar is outside the limits of the directional diagram. Using the linear expansion for the frequency Ω(ϕ, ψ) given by Equation (4.2), as shown in Equation (3.47), ˜ −ϕΩ ˜ −ψΩ ˜ , Ω(ϕ , ψ ) = Ω v 0 h

(4.15)

˜ = Ω (cos ε cos β cos γ − sin ε sin γ ) = Ω cos θ ; Ω 0 max 0 0 0 0 0 max 0

(4.16)

where

Copyright 2005 by CRC Press

1598_book.fm Page 144 Monday, October 11, 2004 1:57 PM

144

Signal and Image Processing in Navigational Systems

˜ = Ω cos ε sin β Ω max 0 0 h

˜ = Ω (cos ε cos β sin γ + sin ε cos γ ) . and Ω v max 0 0 0 0 0 (4.17)

Under the condition ε0 = 0, Equation (4.16) and Equation (4.17) coincide with Equations (3.48) to (3.51). When the variables ϕ and ψ are separable in the function g(ϕ, ψ), then Equation (4.13) can be written in the form of the product that is analogous to Equation (3.52). In this case, the normalized correlation function Rgv ( τ) of the target return signal Doppler ﬂuctuations is different from that given by Equation (3.54), in which we use the function g v2 (ψ ) = gv2 (ψ ) ⋅ S°(β 0 , ψ + γ 0 )sin(ψ + γ 0 )

(4.18)

instead of the function g v2 ( ψ ), and the parameters Ω0 , Ωh , and Ωv in Equations

˜ ,Ω ˜ , and Ω ˜ in Equation (4.16) and (3.48) through (3.51) are replaced with Ω 0 v h Equation (4.17), respectively. Consequently, the total power spectral density of target return signal ﬂuctuations is formed by convolution between two power spectral densities

( )

Sgh (ω ) ≅ gh2 − Ωω ,

(4.19)

h

( )

( ) (

Sgv (ω ) ≅ g v2 − Ωω = gv2 − Ωω ⋅ S° β 0 , γ 0 − v

v

ω Ω v

) sin ( γ

0

−

ω Ω v

)

(4.20)

˜ . and by using the result of the convolution at the frequency ω 0 + Ω 0 Let the directional diagram be determined by the Gaussian distribution law, and the speciﬁc effective scattering area S°(γ) by the exponential law given by Equation (2.150).10,11 Reference to Equation (4.19) and Equation (4.20) shows that Sgh (ω ) ≅ e

2 − π ω2

˜ ∆Ω h

;

),

(4.22)

˜ ˜ ˜ ⋅ ∆( 2 ) = Ω h ,v ⋅ ∆ h ,v . ∆Ω = Ω h ,v h ,v h ,v 2

(4.23)

S (ω ) ≅ e v g

2 k ω − π ω − 1 Ωv 2

∆Ωv

(

(4.21)

⋅ sin γ 0 −

ω Ω v

where

Copyright 2005 by CRC Press

1598_book.fm Page 145 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

145

The convolution between the power spectral densities Sgh (ω ) and Sgv (ω ) of the ﬂuctuations is rigorously determined, but it looks very cumbersome. To obtain a simpler form of convolution between Sgh (ω ) and Sgv (ω ), we have to take into consideration the following circumstance. If the vertical-coverage directional diagram width is not so high in value and the angle γ0 is not so low in value, the generalized directional diagram given by Equation (4.18) is approximately Gaussian:12

g v (ψ ) = g0 ⋅ e

−

π (ψ − ψ 0 )2 ∆(v2 )

.

(4.24)

The generalized vertical-coverage directional diagram has the same effective width ∆v , but it acquires the shift ψ0 , which is low by value, and the coefﬁcient of proportionality g0: 2 π ψ 20

( k + ctg γ 0 )∆ (v2 ) ψ0 = 1 4π

and

g = S°°( γ 0 )sin γ 0 ⋅ e

∆(v2 )

2 0

.

(4.25)

The shift ψ0 consists of two terms. The ﬁrst term is caused by the function S°(γ) given by Equation (2.150) — the speciﬁc effective scattering area. The second term is determined by the target return signal power as a function of the radar range. The power spectral density given by Equation (4.22) takes the following form in this case

S (ω ) ≈ e v g

−π⋅

( ω + δ Ω )2 2 ∆Ω v

(4.26)

where ˜ ⋅ψ = Ω ˜ ⋅ ( k1 + ctg γ 0 )∆ v . δΩ=Ω v v 0 4π (2)

(4.27) ~

In terms of the shift in the frequency of the value of ω0 + Ω0 , the convolution between the power spectral densities Sgh (ω ) and Sgv (ω ), which are determined by Equation (4.21) and Equation (4.26), respectively, gives us the following result:13 )≈e Sg (ω ) = Sgh (ω ) ∗ Sgv (ω ) ∗ δ(ω − ω 0 − Ω 0

Copyright 2005 by CRC Press

−π⋅

( ω − ω 0 − Ω )2 2 ∆Ω

,

(4.28)

1598_book.fm Page 146 Monday, October 11, 2004 1:57 PM

146

Signal and Image Processing in Navigational Systems

˜ 2 = ∆Ω ˜ 2 + ∆Ω ˜ 2 , and Ω = Ω ˜ − δΩ is the average frequency of the where ∆Ω v 0 h power spectral density Sg(ω) of the Doppler ﬂuctuations given by Equation (4.28). The average frequency Ω is different from the average Doppler ~ frequency Ω0 given by Equation (4.16) for the value of δΩ (the error). The meaning and signiﬁcance of the error δΩ can be deﬁned under the condition ε0 = 0.14–16 Then δΩ = Ω 0 ψ 0 tg γ 0 = Ω 0 ⋅

( k1tg γ 0 + 1)∆(v2 ) , 4π

(4.29)

where Ω0 = Ωmax cos β0 cos γ0. The relative value of the error δΩ given by Equation (4.29) has the following form: ( k tg γ 0 + 1)∆(v2 ) δΩ . = δ1 + δ 2 = 1 4π Ω0

(4.30)

The error δ1 is caused by the speciﬁc effective scattering area S°(γ). At k1 = 3.3 and k1 = 13, which corresponds to reﬂection from a plow in summer and rough sea of 1 (see Table 2.1), ∆v = 6° and γ0 = 60°, we obtain the error δ1 equal to 0.5% and 2%, respectively. The error δ2 is less, as a rule. At ∆v = 6°, the error δ2 is equal to 0.1%. The errors δ1 and δ2 depend on the characteristics and parameters of the radar antenna only in the vertical plane — parameters ∆v and γ0. Experimental results regarding the error δ1 as a function of the rough sea are shown in Figure 4.1 at ∆v = 6° and γ0 = 65° when a rough sea is caused by the wind and there is a rippled sea. When there is a rippled sea, the error δ1 is greater because the sea surface becomes smoother in spite of the waves being high, in comparison with the rough sea caused by the wind.17,18 The error δ1 given by the experiment for the Earth’s surface is signiﬁcantly less, as one would expect. At the same values of ∆v and γ0 , the error δ1 given by the experiment is equal to 0.35%, 0.45%, and 0.55% for the forest, ﬁeld, and plow, respectively. These experimental values for the error δ1 coincide very well with theoretical results.19–22 The realistic directional diagram is signiﬁcantly different from its Gaussian directional diagram due to the presence of side lobes. Very often, a changeover from the directional to the Gaussian diagram does not ensure the required accuracy of determination and computer calculation. In this case, the power spectral density of the ﬂuctuations must be deﬁned using the exact approximation for the directional diagram g(ϕ, ψ), taking into consideration the side lobes. For example, see Section 3.2, the numerical integration of the real two-dimensional directional diagram given by the experiment [see Equation (4.13)], or the technique of partial diagrams.23,24 Let us represent the square of the real two-dimensional directional diagram as a sum of partial diagrams of the main beam and side lobes. In doing so, each partial diagram is Gaussian:25

Copyright 2005 by CRC Press

1598_book.fm Page 147 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

147

δ1 (%)

3

2 2

1 Magnitude 1

0

1

2

4

3

FIGURE 4.1 Shift in average Doppler frequency as a function of magnitude of the rough (the curve 1) and rippled (the curve 2) sea.

n

g 2 (ϕ , ψ ) =

∑∑ i=0

− 2π ⋅

m

gij2 ⋅ e

2

[ (ϕ ∆− ϕ ) ij

(2) hij

+

( ψ − ψ ij )2 ∆( 2 ) vij

] ,

(4.31)

j=0

where gij is the relative power of the i-th, j-th side lobe; i is the number of side lobes of the partial diagram in the plane ϕ; j is the number of side lobes of the partial diagram in the plane ψ; ϕij is the angle coordinate of the center of the side-lobe, relatively the center of the main beam in the plane ϕ; ψij is the angle coordinate of the center of the side lobe, relatively the center of the main beam in the plane ψ; ∆hij is the effective width of the i-th, j-th side lobe in the plane ϕ; ∆vij is the effective width of the i-th, j-th side lobe in the plane ψ; i = j = 0 is the case of the main beam. All these parameters are determined using the experimental two-dimensional directional diagram. We can use the width at the level 0.5 from the maximum instead of the effective width of the main beam and side lobes. We can assume that the value of S°(γ) is constant within the limits of the side lobe and do not consider changes in the radar range.26,27 Each partial diagram can be considered independent of other partial diagrams. Consequently, the target return signals for partial diagrams are noncoherent. In this case, the total power spectral density of the ﬂuctuations is equal to sum of the independent partial power spectral densities formed by individual side lobes. Using Equation (4.28) and taking into account the contribution in the energy of each side lobe, we can write the power spectral density in the following form:28,29

Copyright 2005 by CRC Press

1598_book.fm Page 148 Monday, October 11, 2004 1:57 PM

148

Signal and Image Processing in Navigational Systems

n

S(ω ) =

m

n

m

∑ ∑ S (ω) =∑ ∑ ∆Ω pij

⋅e

ij

i=0

j=0

i=0

j=0

−π⋅

( ω − ω 0 − Ωij )2 ∆Ωij2

,

(4.32)

ij

where S00(ω) is the power spectral density for the main beam (the main lobe);

pij =

Pλ 2 gij2 ∆ hij ∆ vij sin γ ij S°( γ ij ) 128 π 3 h2

;

(4.33)

Ω ij = 4 πVλ−1 cos β ij cos γ ij ;

(4.34)

∆Ω ij = 2 2 π Vλ−1 ( ∆ hij sin β ij )2 + ( ∆ vij cos β ij sin γ ij )2 ;

(4.35)

βij = β0 + ϕij; and γij = γ0 + ψij.

4.3

The Pulsed Searching Signal with Stationary Radar

4.3.1

General Statements

At V = 0, we obtain that Ω(ϕ, ψ) = 0 and µ = 1. Introducing a new variable [see Equation (2.142)] z=t−

2 ρ( ψ ) c

and

c* dz = c dψ

(4.36)

in Equation (4.5), we can write ∞

R en (t , τ) =

∑ R (t, τ − nT ) , 0

p

(4.37)

n= 0

where

∫

R0 (t , τ) = p* P ( z − 0.5τ) ⋅ P * ( z + 0.5τ) ⋅ g v2 ( ψ * + c* z) dz

(4.38)

is the correlation function at n = 0; p* =

Copyright 2005 by CRC Press

PG02 λ 2 ∆ (h2 ) c*S°( γ * )sin γ * ; 64 π 3 h2

(4.39)

1598_book.fm Page 149 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

149

c* = 0.5cρ–1tg γ*; and c*z = ψ – ψ*. Here, a dependence on the time parameter t — a feature of the nonstationary state — is changed by dependence on the angles ψ*(t) and γ*(t) = ψ*(t) + γ0. The angles ψ*(t) and γ*(t) are related to the time parameter t by the formula [see Equation (2.139)]: t=

2 ρ* 2h 2h = = . c c sin γ * c sin(ψ * + γ 0 )

(4.40)

The correlation function of the ﬂuctuations given by Equation (4.37) is a periodic function of the parameter τ with the period Tp (see Figure 3.1a, the solid line). This correlation function deﬁnes the rapid ﬂuctuations in the radar range. Because, in the general case, the variables t(ψ*) and τ are not separable, the correlation function is not separable either — the spectral characteristics depend on the time parameter t. The corresponding instantaneous power spectral density of the ﬂuctuations is a regulated function.30,31 The envelope of the power spectral density is the Fourier transform of the envelope of the correlation function given by Equation (4.38) and also depends on the time parameter t. The instant of time t (or the angle ψ*) deﬁnes the interval within the limits of which the variable ψ* + c*z in the integrand function gv(ψ* + c*z) (see Figure 4.2) changes, and the form of the total part of the product of the functions P(t) and P*(t) (see Figure 4.2, the hatched area). Thus, we can say that the shape of the waves of the correlation function given by Equation (4.38) and the envelope of the regulated power spectral density are deﬁned by the instant of time t. Naturally, the target return signal power also depends on the instant of time t. S(t − 2ρ c ) P(t − 2ρ + 0.5 τ) c

P(t − 2ρ − 0.5 τ) c

gv(ψ∗ + c∗z)

2ρ c 0

t (ψ∗)

FIGURE 4.2 The instantaneous power spectral density of the target return signal ﬂuctuations as a function of the delay t.

Copyright 2005 by CRC Press

1598_book.fm Page 150 Monday, October 11, 2004 1:57 PM

150

Signal and Image Processing in Navigational Systems

The instantaneous power spectral density of the ﬂuctuations in the radar range is presented in Figure 4.3 as a function of time t in the case of the individual pulsed searching signal propagated along the surface of the twodimensional target — the continuous power spectral density. The analogous power spectral density of the ﬂuctuations in the case of the inﬁnite periodic sequence of pulsed searching signals is shown in Figure 4.4 as a function of the time-cross-section t(ρ*) — the regulated power spectral density of the ﬂuctuations [see Equation (2.1)]. S(ω) ω

ω0

t 0

FIGURE 4.3 The instantaneous power spectral density of the target return signal ﬂuctuations with the individual pulsed searching signal.

S(ω) ω

ω0

t (ρ∗) 0

FIGURE 4.4 The instantaneous power spectral density of the target return signal ﬂuctuations with a sequence of pulsed searching signals.

Copyright 2005 by CRC Press

1598_book.fm Page 151 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

151

Reference to Equation (4.38) shows that if the vertical-coverage directional diagram is wide, i.e., the condition ∆v >> ∆p (or τp << Tr) is true, we can assume that gv(ψ* + c*z) ≈ gv(ψ*). In this case, we can write

∫

R0 (t , τ) = p* g v2 ( ψ * ) P( z − 0.5τ) ⋅ P * ( z + 0.5τ) dz .

(4.41)

This correlation function of ﬂuctuations in the radar range is equivalent to that given by Equation (3.3) for the three-dimensional (space) target under the condition µ = 1, i.e., the radar is stationary. Consequently, all results obtained in Section 3.1 are true for the case considered here. However, it differs from the case discussed in Chapter 3 by the other dependence between the target return signal power and time t (or radar range) that was discussed in Section 2.4 and Section 2.5. In the discussed approximation procedure, the variables t(ψ*) and τ are separable, as shown in Equation (4.41). Consequently, the correlation function of the ﬂuctuations is separable. Let us consider and discuss some examples in practice.

4.3.2

The Arbitrary Vertical-Coverage Directional Diagram: The Gaussian Pulsed Searching Signal

Reference to Equation (4.37) shows that if the pulsed searching signal is Gaussian [see Equation (2.105)], we can write ∞

R en (t , τ) = p(t) ⋅

∑e

−π⋅

( τ − nTp )2 2 τ 2p

,

(4.42)

dz .

(4.43)

n= 0

where ∞

p(t) = p*

∫ g (ψ 2 v

*

+ c * z) ⋅ e

2 − 2 π 2z

τp

−∞

Here the variables t and τ are separable, i.e., we can say that the correlation function of the target return signal ﬂuctuations is a separable process for any shape and width of the vertical-coverage directional diagram. The power p(t) of the target return signal as a function of the parameter t for the case of the Gaussian vertical-coverage directional diagram was discussed in Section 2.5. The normalized correlation function of the ﬂuctuations following from Equation (4.42) coincides with that given by Equation (3.20) in the case of the three-dimensional (space) target when the radar is stationary, µ = 1. This correlation function is represented as a comb of Gaussian waves with the effective width equal to 2τ p . The corresponding power spectral Copyright 2005 by CRC Press

1598_book.fm Page 152 Monday, October 11, 2004 1:57 PM

152

Signal and Image Processing in Navigational Systems

density is the regulated function, and the envelope of the power spectral density is Gaussian with the effective bandwidth equal to ( 2 τ p ) −1 . The envelope of the power spectral density of the ﬂuctuations is similar to that given by Equation (3.21) if we replace the parameter τ′p with the parameter τp in Equation (3.21). 4.3.3

The Arbitrary Vertical-Coverage Directional Diagram: The Square Waveform Pulsed Searching Signal

The directional diagram is deﬁned by continuous functions with continuous derivatives, and we can use the Taylor-series expansion32–34 ∞

gv2 (ψ * + c* z) =

∑a z

k

k

,

(4.44)

k=0

ck g2 k ( ψ )

where a0 = g v2 ( ψ * ) and ak = * vk! * . In the case of the square waveform pulsed searching signal, reference to Equation (4.38) and Equation (4.44) shows that all terms with the odd powers of the value z are equal to zero, and we can write 0.5( τ p − τ )

∫

R0 (t , τ) = p*

∞

gv2 (ψ ∗ + c∗ z) dz = p′

∑ b (1 − k

k=0

− 0.5( τ p − τ )

|τ | τp

) 2 k +1 ,

(4.45)

where p ′ = p* τ p g v2 ( ψ * ) ; b0 = 1; bk =

a2 k τ 2p k 2 (2 k + 1)!g (ψ * ) 2k

2 v

∆ p = c* τ p =

=

2k ∆ (p )

[ gv2 (ψ * )] 2 k ; 2 (2 k + 1)! gv2 (ψ * ) 2k

cτp 2 ρ* ctg γ *

=

⋅

c τ p sin 2 γ * 2 h cos γ *

(4.46)

(4.47)

is the angle dimension of the resolution element in the radar range [see Equation (2.140)]. Because the coefﬁcients bk depend on the time t through the parameters ∆p and gv(ψ*), the variables t and τ are not separable in the general case, and thus the correlation function of the target return signal ﬂuctuations is not separable. As the delay or radar range ρ* is increased, i.e., the parameter γ* is decreased, the coefﬁcients bk are rapidly decreased, and the inﬂuence of the high-order terms so quickly weakened that the duration of the pulsed

Copyright 2005 by CRC Press

1598_book.fm Page 153 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

153

searching signal is short. In the pulsed searching signals with very short duration, or with low values of γ*, we need consider only the ﬁrst term. Then R en (t , τ) ≈ p ′(1 − |Tτp|) .

(4.48)

For this approximation, the correlation function is separable. The instantaneous normalized correlation function and power spectral density are the same as in the case of scanning the three-dimensional (space) target with the stationary radar, µ = 1 [see Equation (3.18) and Equation (3.19)]. In rigorous analysis, when it is necessary to consider high-order terms in Equation (4.45), the envelope of the power spectral density of the ﬂuctuations can be written in the following form35 ∞

Spen (ω , t) ≅

∑ b (ψ ) ⋅ S (ω) , k

*

(4.49)

k

k=0

where

Sk (ω ) = (−1)

k +1

⋅

2 ( 2 k + 1)! ( ωτ p )2 k + 2

k

[

⋅ cos ωτ p −

∑ (−1)

m

⋅

( ωτ p )2 m ( 2 m )!

]

and

Sk (0) =

1 k +1

.

m= 0

(4.50) We can write the power spectral densities at k = 0, 1, 2 in the following form: S0 (ω ) = sinc 2 (0.5ωτ p )

and S0 (0) = 1 ;

(4.51)

S1 (ω ) = 6(ωτ p ) −2 [1 − S0 (ω )]

and

S1 (0) = 0.5 ;

(4.52)

S2 (ω ) = 10(ωτ p ) −2 [1 − 2S1 (ω )]

and

S2 (0) ≈ 0.33 .

(4.53)

These power spectral densities are shown in Figure 4.5. The effective bandwidth of the k-th power spectral density increases with an increase in the number: τp

∆Fk =

∫ (1 −

− τp

Copyright 2005 by CRC Press

|τ | τp

) 2 k + 1 dτ = kτ+ 1 . p

(4.54)

1598_book.fm Page 154 Monday, October 11, 2004 1:57 PM

154

Signal and Image Processing in Navigational Systems

1.0

S(ω)

0.8 0.7 S 0 (ω)

0.6 S1 (ω) 0.4 S (ω) 2 0.2 0.1 −9 −8 −7 −6 −5 −4 −3 −2 −1

ω τp 0 1

2

3

4

5

6

7

8

9

FIGURE 4.5 The components S0(ω), S1(ω), and S2(ω) of the power spectral density of the target return signal ﬂuctuations in the radar range (the solid line) and their sum (the dotted line) at ψ* = 0° and ∆*p ∆h

= 0.5.

Because the contribution of these terms depends on the value of the current angle γ*, the instantaneous power spectral density is a function of the delay of the target return signal or radar range ρ*. 4.3.4

The Gaussian Vertical-Coverage Directional Diagram: The Square Waveform Pulsed Searching Signal

Reference to Equation (4.45) shows that R0 (t , τ) =

π 2

{ { [ψ

⋅ p∗ Φ

2π ∆v

∗

+

∆p 2

(1 − |ττ|)]} − Φ { ∆2 π [ψ ∗ − ∆2

p

p

v

}} ,

(1 − |ττp|)]

(4.55) where Φ(x) is the error integral given by Equation (1.27). Determination of the correlation function of the ﬂuctuations given by Equation (4.55) is true for any relationship between ∆v and ∆p. Further, this determination does not have any limitations with regard to the width of the vertical-coverage directional diagram and the angle ψ*. These values can be high.36 When the width of the vertical-coverage directional diagram is low in value, and the value of the angle γ0 is not so low, we can use the linear relationship between the variables t and ψ* [see Equation (2.142)], and the values of the angle ψ*, ∆v , and ∆p in Equation (4.55) can be deﬁned using the variables Tr , t – Td , and τp:

Copyright 2005 by CRC Press

1598_book.fm Page 155 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target ψ k Td − t ≈ ∆v 2 Tr ∆p ∆v

at

τp

≈

at

2 Tr

155

ψ * ctg γ 0 << 1 ;

(4.56)

ρ* ctg γ * ≈ ρ0 ctg γ 0 ,

(4.57)

where Tr =

Td ∆ v 2

⋅ ctg γ 0

(4.58)

is the effective duration of the target return signal [see Equation (2.151)] and Td =

2 ρ0 2h = c c sin γ 0

(4.59)

is the delay of the target return signal from the scatterer placed on the axis of the directional diagram in the angle γ0. In this case, the correlation function of the ﬂuctuations given by Equation (4.55) can be written in the explicit functional form of the parameter t: R0 (t , τ) ≅

π 2

{ { [t − T

⋅ p* Φ

π Tr

d

+

τp 2

(1 − |ττ|)]} − Φ { Tπ [t − Td − τ2 (1 − |ττ|)]}} . p

p

r

p

(4.60) Equation (4.55) can be represented in the series expansion form [see Equation (4.45)], where ∆

bk = H 0 ( x) = 1 ;

2k ( π2 )k ∆ v ( 2 k +1)!

( p)

ψ

⋅ H 2 k ( x)

and

x = 2 π ⋅ ∆ v* ;

H 2 ( x) = − 2 + 4 x 2 ;

and

H 4 ( x) = 12 − 48 x 2 + 16 x 4 (4.62)

(4.61)

are the Hermite polynomials. Because the coefﬁcients bk depend on the variable ψ*(t), the variables t and τ are not separable in the general case, and the shape of the envelope of the instantaneous power spectral density is changed during the propagation of the target return signal. One can see from Equation (4.61) that the ratio plays the main role in this formula. If the ratio

Copyright 2005 by CRC Press

∆p ∆v

∆p ∆v

is so low in value and

1598_book.fm Page 156 Monday, October 11, 2004 1:57 PM

156

Signal and Image Processing in Navigational Systems

for any values of the angle ψ*(t) we can neglect all forms of the series expansion except the ﬁrst term, the correlation function of the ﬂuctuations takes the form as in Equation (4.48) and is independent of the time variable t. In this approximation, the correlation function is separable. For example, at γ0 = 60°, τp = 1 µsec, h = 3000 m, ∆v = 6°, we obtain the ratio i.e., it is not so low in value. In the condition

∆p ∆v

∆p ∆v

= 0.75,

> 1, the series given by

Equation (4.45) converges slowly. At ψ* = 0 or t = Td, Equation (4.55) can be simpliﬁed and has the following form: R0 ( τ , Td ) = π ⋅ Φ

[

π 2

∆

]

⋅ ∆ vp (1 − |ττp|) ,

(4.63)

where

∆ p =

c τ p sin 2 γ 0 2 h cos γ 0

.

(4.64)

The coefﬁcients bk given by Equation (4.61) take the following form:

bk =

∆p 2k ( π2 )k ∆ v ( 2 k +1)!

( )

⋅ H 2 k ( 0) =

( − π )k ( 2 k )!!( 2 k +1)

∆

⋅ ( ∆ vp )2 k .

(4.65)

The normalized correlation function of the ﬂuctuations given by Equation (4.63) is shown in Figure 4.6 at various values of the ratio

∆p ∆v

. The periodic

normalized correlation function given by Equation (4.37), with the waves given by Equation (4.63), is shown in Figure 4.7 at various values of the duration τp of the pulsed searching signal and at the constant value of ∆v . Under the condition

∆p ∆v

<<1, the shape of waves of the periodic normalized

correlation function is close to the triangular form [see Equation (4.48)]. As the duration τp of the pulsed searching signal is increased, the waves of the periodic normalized correlation function are naturally expanded. The width of the waves — the base of the triangular form — is always equal to 2τp and the apex of waves is smoothed, but the duration and slope of the leading and trailing edges of the pulsed searching signal are the same. In the condition τp → Tp , the waves of the periodic normalized correlation function of the ﬂuctuations are merged in line with the constant level. This means that the intraperiod ﬂuctuations are absent in passing from the pulsed searching signal to the continuous searching signal. The physical meaning of this characteristic of the correlation function is clear. When the pulsed Copyright 2005 by CRC Press

1598_book.fm Page 157 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

157

R0 (τ,Td)

1.0

5 3 4 2 1

0.5

τ τp −1.0

−0.5

0

0.5

1.0

FIGURE 4.6 The normalized correlation function of the target return signal ﬂuctuations in the radar range for individual waves at the stationary radar, ψ* = 0° and τp = const: (1) ∆p ∆h

= 1.0; (4)

∆p ∆h

= 1.5; (5)

∆p ∆h

∆p ∆h

= 0; (2)

∆p ∆h

= 0.5; (3)

= 2.0.

R en (t, τ)

6 6 1 2 0

3

4

5

5

4

3

2 1

τ Tp

FIGURE 4.7 The normalized correlation function of the target return signal ﬂuctuations in the radar range for a sequence of waves at the stationary radar, ψ* = 0° and τp = const: (1) and (2) ∆ p < ∆v ; (3) and (4) ∆ p ≈ ∆v ; (5) and (6) ∆ p > ∆ v..

searching signal duration is high in value, i.e., the pulsed searching signals completely overlap the scanning surface of the two-dimensional target, the resulting pulsed target return signal assumes a shape that is very similar to that of the pulsed searching signal (see Figure 4.8a). In this case, there are the ﬂuctuations of the slope of the leading and trailing edges, which are formed in scanning the two-dimensional (surface) target (see Figure 4.8b). The top of the pulsed target return signal is ﬂat and the ﬂuctuations are absent. The power spectral density given by Equation (4.49) corresponds to the sum of the terms in Equation (4.50)–Equation (4.53) and is shown in Figure 4.5 by the dotted line in the conditions ψ* = 0° and

∆p ∆v

= 0.5.

However, in spite of the deﬁnition of the envelope of the correlation function given by Equation (4.55) being formally true for any values of ∆op

Copyright 2005 by CRC Press

1598_book.fm Page 158 Monday, October 11, 2004 1:57 PM

158

Signal and Image Processing in Navigational Systems

(a)

t

(b)

t FIGURE 4.8 (a) Shape of the pulsed searching signal, and (b) shape of the target return signal at ∆°p >> ∆v .

and ∆v , we should be very careful in using Equation (4.55) at the condition

∆p ∆v

> 1, because the stochastic process degenerates when the duration

of the pulsed searching signal is very high in value. Within the limits of the time interval in which the pulsed searching signal completely covers the scanning surface of the two-dimensional target, we cannot assume that the target return signal is a Poisson stochastic process, i.e., we cannot consider the target return signal to be the sum of the elementary signals arising at the random instants of time. As mentioned previously, in the condition

∆p ∆v

>> 1, the pulsed target return signal has a ﬂat top because the

propagation of the pulsed searching signal along the surface of the twodimensional target is not accompanied by initiation of new elementary signals during the large time intervals. In other words, we can say that the stochastic process is converted from the nonsingular process, as in the case of the pulsed searching signal with the short duration, into the two-parametric singular process, for example, A cos (ω t + ϕ), in which only the parameters A and ϕ are stochastic.37 The correlation function of this process does not completely deﬁne the properties of the process because it has not ceased to exist as a Gaussian stochastic process.

Copyright 2005 by CRC Press

1598_book.fm Page 159 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target S (ω) Tp

0 ω0

159

t

ω ω 0 + nΩp

FIGURE 4.9 The instantaneous power spectral density of the target return signal ﬂuctuations as a function of time under the condition ∆°p << ∆v .

Under the condition

∆p ∆v

>>1, in the act of the pulsed searching signal over-

lapping the scanned surface of the two-dimensional target, the ﬂuctuations have a very short correlation interval at the start. Thereafter, it increases so that the ﬂuctuations are absent at the top of the pulsed target return signal. As the pulsed searching signal is run down from the scanned surface of the two-dimensional target, the process goes into inverse sequence: in the beginning, the slow target return signal ﬂuctuations arise, and thereafter, they become rapid (see Figure 4.8b). The instantaneous power spectral density as a function of time t corresponding to the process described in the preceding text is shown in Figure 4.9 under the condition ω >> ω0 , where it is symmetric with respect to the frequency ω0 . In the beginning, the instantaneous power spectral density has an effective bandwidth high in value and low power. Thereafter, the effective bandwidth is decreased and the power is increased. At the end of the pulse, the spectral density is expanded again — i.e., the effective bandwidth is increased — and the power is decreased. Because the process is rigorous and periodic, i.e., the radar is stationary and the interperiod ﬂuctuations are absent, we have the regulated power spectral density with the distance Ωp between harmonics. The effective bandwidth in the stationary region is deﬁned by the duration τp of the pulsed searching signal: ∆F ≈ (τp)–1.

Copyright 2005 by CRC Press

1598_book.fm Page 160 Monday, October 11, 2004 1:57 PM

160

Signal and Image Processing in Navigational Systems

4.4

The Pulsed Searching Signal with the Moving Radar: The Aspect Angle Correlation Function

4.4.1

General Statements

The aspect angle correlation function of the target return signal ﬂuctuations is determined by Equation (4.8). Introducing a new variable z given by Equation (4.36), we can write ∞

Rγ (t , τ) = N

∑ ∫ P [z − 0.5(µτ − nT )] ⋅ P [z + 0.5(µτ − nT )] , *

p

p

n= 0

× gv2 (ψ * + c* z) ⋅ e

− jc*Ω γ zτ

(4.66)

dz

where c* = 2ρc * tg γ*; c*z = ψ – ψ*; and Ωγ is determined by Equation (4.10). The aspect angle normalized correlation function given by Equation (4.66) has a very complex structure and will be investigated in more detail using speciﬁc examples. Here we study only the interperiod ﬂuctuations. We can simplify some formulae in Equation (4.9) and Equation (4.10) that containing the parameters Ωβ and Ωγ , assuming that the radar moves only in the horizontal plane, i.e., ε0 = 0. We can always omit this limitation. The normalized correlation function in the glancing radar range follows from Equation (4.66) at the condition τ = nTp:

∫

Rγ (t , τ) = N Π 2 ( z) ⋅ g v2 ( ψ * + c* z) ⋅ e

− jc∗Ω γ zτ

dz .

(4.67)

The corresponding power spectral density takes the following form:

(

Sγ [ω , ψ * (t)] ≅ Π2 −

ω c∗Ω γ

) ⋅ g (ψ − ) . 2 v

*

ω Ωγ

(4.68)

Reference to Equation (4.68) shows that the power spectral density depends essentially on the angle ψ* because the parameters c* and Ωγ depend on it. The relative position of the target return signal and the vertical-coverage directional diagram also depends on the angle ψ* because the normalized correlation function given by Equation (4.67) is not separable, i.e., the variables t and τ are not separable. When the vertical-coverage directional diagram width — the beam width — is large in value, i.e., the condition ∆v >> ∆p is satisﬁed, the shape of the power spectral density can be determined in the following form:38

(

Sγ (ω , ψ * ) ≅ Π2 −

Copyright 2005 by CRC Press

ω c*Ω γ

).

(4.69)

1598_book.fm Page 161 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

161

The effective power spectral density bandwidth is determined as follows: ∆Ωτ = kp τ p c* ⋅ Ω γ = ∆ p ⋅ Ω γ =

2 π kp c τ pV sin 3 γ * cos β 0 λ h cos γ *

.

(4.70)

The effective bandwidth ∆Ωτ depends essentially on the aspect angle γ* (see Figure 4.10, the solid line; the dotted line will be discussed in Section 10.2). With high values of the aspect angle γ*, the effective bandwidth ∆Ωτ is very high. For example, at V = 300

m sec

, l = 3cm, τp = 1µsec, γ* = 45°, β0 = 45°, we ∆Ω

obtain the effective bandwidth ∆Fτ = 2 πτ = 350 Hz. Under the condition γ* → 90°, both Equation (4.70) and Equation (2.140) are not true. If the condition h = const is true, with an increase in the radar range ρ*, the effective bandwidth ∆Ωτ decreases sharply and tends to approach zero because the angle ∆p given by Equation (2.140) is decreased. Deviation of the radar antenna from the direction of the moving radar (β0 ≠ 0) leads to decrease in the effective bandwidth ∆Ωτ of the power spectral density. If the duration of the pulsed searching signal is high in value, i.e., the condition ∆p >> ∆v is satisﬁed, then, for those time cross sections in which the pulsed searching signal completely overlaps the scanned surface of the two-dimensional target, the shape of the power spectral density is deﬁned by the shape ∆Ωτ ∆Ωτ|γ = π/4 *

5

4

3

2

1

γ* 0

20°

40°

60°

80°

FIGURE 4.10 The bandwidth ∆Ωτ of the power spectral density of the target return signal ﬂuctuations as a function of the aspect angle γ*: ∆Ωτ — the solid line; |∆Ωτ – ∆Ωω |— the dotted line.

Copyright 2005 by CRC Press

1598_book.fm Page 162 Monday, October 11, 2004 1:57 PM

162

Signal and Image Processing in Navigational Systems

of the square of the vertical-coverage directional diagram under the condition ψ* = 0 or γ* = γ0:

( ).

Sγ (ω , ψ ∗ ) ≅ gv2 −

ω Ωγ

(4.71)

Equation (4.71) is equivalent to Equation (3.55) and Equation (4.20) if the conditions S°(β0, ψ + γ0) ≈ S°(β0, γ*) and sin(ψ + γ0) ≈ sin γ* are satisﬁed in Equation (4.20) and ε0 = 0. The effective power spectral density bandwidth of the interperiod ﬂuctuations given by Equation (4.71) has the following form: ∆Ω v = ∆(v2 ) ⋅ Ω γ =

4 π V (2) ⋅ ∆ v cos β 0 cos γ 0 λ

(4.72)

and coincides with Equation (3.84). The effective bandwidth given by Equation (4.72) is different from that given by Equation (4.70). We use the width ∆(v2 ) of the vertical-coverage directional diagram, i.e., the beam width, in Equation (4.72) instead of the angle ∆p used in Equation (4.70). The normalized correlation function in the ﬁxed radar range can be obtained based on Equation (4.66) under the condition τ = nTp. 4.4.2

The Arbitrary Vertical-Coverage Directional Diagram: The Gaussian Pulsed Searching Signal

If the pulsed searching signal is Gaussian [see Equation (2.105)], reference to Equation (4.66) shows that Rγ ( τ , ψ * ) = Rγ′ ( τ) ⋅ Rγ′′( τ , ψ * ) ,

(4.73)

where ∞

Rγ′ (τ) =

∑e

−π ⋅

( µτ − nTp )2 2 τ 2p

;

(4.74)

n= 0

∫

Rγ′′( τ , ψ ∗ ) = N g ( ψ * + c* z) ⋅ e 2 v

2 − 2 π2z − jc* Ω γ zτ

τp

dz .

(4.75)

The normalized correlation function Rγ′ ( τ) of the interperiod ﬂuctuations has a comb structure with the same Gaussian waves, with the effective bandwidth equal to 2 τ ′p = 2 τpµ–1 and the period equal to Tp′ = Tpµ–1. Rγ′ ( τ) coincides with the normalized correlation function Rp(τ) in the radar range

Copyright 2005 by CRC Press

1598_book.fm Page 163 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

163

scanning the three-dimensional (space) target [see Equation (3.6) and Equation (3.20); Figure 3.1a, the dotted line]. The corresponding power spectral density Sγ′ (ω) is the regulated function given by Equation (3.15) and Equation (3.21) (see Figure 3.1b, the dotted line). The normalized correlation function Rγ′′ (τ, ψ*) deﬁnes the slow ﬂuctuations caused by the difference in the Doppler frequencies in the aspect-angle plane. Rγ′′ (τ, ψ*) characterizes the destruction of correlation from period to period because, under the condition τ = nTp′, the aspect-angle normalized correlation function Rγ(τ, ψ*) is deﬁned by the normalized correlation function Rγ′′ (τ, ψ*) in the aspect angle plane: Rγ(τ, ψ*) = Rγ′′ (τ, ψ*). In other words, Rγ′′ (τ, ψ*) is the normalized correlation function in the glancing radar range. Thus, with the Gaussian pulsed searching signal, the aspect-angle normalized correlation function given by Equation (4.73) and the total normalized correlation function are deﬁned by the product of the intraperiod and interperiod ﬂuctuations, as in the case of scanning the three-dimensional (space) target. The power spectral density corresponding to the normalized correlation function Rγ′′ (τ,ψ*), given by Equation (4.75), can be written in the following form: Sγ′′(ω , ψ * ) ≅ g (ψ * − 2 v

ω Ωγ

)⋅ e

2 − π⋅ ω 2

∆Ω τ

,

(4.76)

where the effective bandwidth ∆Ωτ is determined by Equation (4.70) under the condition kp = ( 2 ) −1 . The power spectral density Sγ′′ (ω, γ*) of interperiod ﬂuctuations is the particular case of Sγ[ω, ψ*(τ)] in the glancing radar range given by Equation (4.68). When the vertical-coverage directional diagram width is large in value, i.e., the condition ∆v >> ∆p is true, Sγ′′ (ω, ψ*) takes the following form: Sγ′′(ω , ψ * ) ≅ e

2 − π⋅ ω 2

∆Ω τ

.

(4.77)

This formula is a particular case of the power spectral density given by Equation (4.69). If the duration of the pulsed searching signal is high in value, i.e., the condition ∆p >> ∆v is satisﬁed, and at the values of ψ* in which the pulsed searching signal completely overlaps the scanned surface of the twodimensional target, the power spectral density of the interperiod ﬂuctuations is determined by Equation (4.71). The power spectral density corresponding to the normalized correlation function given by Equation (4.73) is deﬁned by the convolution between Sγ′ (ω) and Sγ′′ (ω, ψ*) (see Figure 3.3b), ∞

Sγ (ω , ψ * ) ≅

∑ S′′(ω − nΩ′ , ψ ) ⋅ e γ

n= 0

Copyright 2005 by CRC Press

p

*

 nΩ′p  −π ⋅ Ω′p   ∆Ω

2

,

(4.78)

1598_book.fm Page 164 Monday, October 11, 2004 1:57 PM

164

Signal and Image Processing in Navigational Systems

where ∆Ω ′p = 2 π ⋅ ∆Fp′ = 2 π( τ ′p ) −1 .

(4.79)

Consider the interperiod ﬂuctuations in the ﬁxed radar range, assuming that the condition τ = nTp in Equation (4.73)–Equation (4.75) is true. Then

Rγ′ (τ) = e

 µ τ − π ⋅  2  τp 

2

(4.80)

,

where µ = µ − 1 = 2Vc −1 cos β 0 cos γ * .

(4.81)

The correlation interval of the normalized correlation function in the ﬁxed radar range given by Equation (4.80) is very high in value: τ ′c = ( ∆F ′)−1 =

cτp 2 V cos β 0 cos γ *

.

(4.82)

The correlation interval is deﬁned by the time required for the radar moving with the velocity V to travel the distance equal to

c τp 2 cos β 0 cos γ *

. Compared with

Equation (3.157), that is equal to the effective bandwidth of the scanned element resolved in the radar range at the cross section, which is parallel to the direction of the moving radar. Within the limits of the time interval equal to the correlation length τ′c , all scatterers ﬁlling the scanned element resolved in the radar range are exchanged. The power spectral density in the ﬁxed radar range corresponding to the normalized correlation function given by Equation (4.80) has the following form: Sγ′ (ω ) ≅ e

−π⋅

(

ω 2 π ∆F ′

)

2

.

(4.83)

The effective bandwidth ∆F′ = ( τ ′c ) −1 given by Equation (4.95) is very low in value, as a rule. The total power spectral density in the ﬁxed radar range corresponding to the normalized correlation function given by Equation (4.73) is deﬁned by the convolution between Sγ′ (ω ) and Sγ′′(ω , ψ ∗ ) , given by Equation (4.83) and Equation (4.76), respectively.

Copyright 2005 by CRC Press

1598_book.fm Page 165 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target 4.4.3

165

The Gaussian Vertical-Coverage Directional Diagram: The Gaussian Pulsed Searching Signal

If the vertical-coverage directional diagram is Gaussian, Equation (4.73) is true when Rγ′′( τ , ψ * ) = e − π ( ∆F ′′τ ) + jΩ′′τ , 2

(4.84)

where Ω ′′ =

∆F ′′ =

ψ *Ωγ ∆(v2 )

1+

∆Fτ ⋅ ∆F ∆F + ∆F 2 τ

;

(4.85)

∆(p2 )

2 v

=

∆Fτ 1+

∆(p2 )

.

(4.86)

∆(v2 )

The power spectral density of the interperiod ﬂuctuations, which can be obtained by the Fourier transform of the normalized correlation function given by Equation (4.84) or follows from Equation (4.76), takes the following form Sγ′′(ω , ψ * ) ≅ e

2 − π ⋅ ( ω − Ω ′′ 2)

( ∆Ω ′′ )

,

(4.87)

where Ω″ and ∆Ω″ are determined by Equation (4.85) and Equation (4.86), respectively. Reference to Equation (4.86) shows that under the condition ∆p << ∆v , the effective bandwidth ∆F″ is close to that given by Equation (4.70). If the condition ∆p >> ∆v is true, ∆F″ is close to that given by Equation (4.72). The effective bandwidth ∆F″ is a complex function of the angle γ* because both the effective bandwidth ∆Fτ and the angle dimension ∆p of the resolved element depend on the angle γ* .39 The ratio

∆p ∆v

is also a function of the angle γ* . With the same parameters

of the radar antenna τp , λ, ∆v and the ﬁxed conditions of radar motion, for example, V and h, the condition ∆p >> ∆v can be satisﬁed when the radar range is low in value — the aspect angle is high in value — and if the condition ∆p << ∆v would be satisﬁed, the radar range is high in value. The average frequency Ω″ of the power spectral density Sγ′′ (ω, ψ*) is proportional to the angle ψ* = γ* – γ0 and depends on the parameters Ωγ and ∆p. In other words, the average frequency Ω″ is a function of the angle γ*(t). On the axis

Copyright 2005 by CRC Press

1598_book.fm Page 166 Monday, October 11, 2004 1:57 PM

166

Signal and Image Processing in Navigational Systems

of the directional diagram, we have the equalities ψ* = 0° and Ω″ = 0. Therefore, we can write  Ω ′′ = ψ *Ωγ ∆(p2 ) ≅ 0  ∆(v2 )   Ω ′′ ≈ ψ * Ω γ 

∆ p << ∆ v ;

at

(4.88)

∆ p >> ∆ v .

at

The total aspect-angle normalized correlation function of the interperiod ﬂuctuations is deﬁned by the product of the normalized correlation functions given by Equation (4.73) and Equation (4.84) and can be written in the following form: ∞

Rγ′′(τ , ψ * ) = e

− π ( ∆F′′τ )2 + jΩ′′τ

⋅

∑e

−π⋅

( µτ − nTp )2 2 τ 2p

.

(4.89)

n= 0

The power spectral density corresponding to the total normalized correlation function Rγ(τ, ψ*) is determined in the following form according to Equation (4.78) and Equation (4.87) ∞

Sγ (ω ) ≅

∑e

 nΩ′p  −π ⋅  ∆Ω′p 

2

⋅e

 ω − Ω′′ − nΩ′p  −π ⋅ ∆Ω′′  

2

,

(4.90)

n= 0

where the effective bandwidth ∆Ω′p is determined by Equation (4.79), and the effective bandwidth ∆Ω″ can be written in the form ∆Ω″ = 2π∆F″. The power spectral density given by Equation (4.90) has a comb structure, in which the waves at the frequencies ω = nΩ ′p + Ω″ have the effective bandwidth ∆Ω″. The envelope of the waves is Gaussian with the effective bandwidth ∆Ω′p . As we noted in Section 4.4.2, Rγ′′ (τ, ψ*) and Sγ′′ (ω, ψ*) are the normalized correlation function and the power spectral density in the glancing radar range. Rγ′′ (τ, ψ * ) and Sγ′′ (ω, ψ * ) are completely deﬁned by Equation (4.84)–Equation (4.87). The normalized correlation function of the interperiod ﬂuctuations in the ﬁxed radar range is deﬁned by the product of the normalized correlation functions Rγ′ (τ) and Rγ′′ (τ, ψ*) given by Equation (4.80) and Equation (4.84), respectively, and can be written in the following form: 2

Rγ (τ , ψ * ) = e − π[( ∆F′′τ )

+ ( ∆F′τ )2 ] + jΩ′′τ

.

(4.91)

The power spectral density is determined by Equation (4.87) with the effective bandwidth given by

Copyright 2005 by CRC Press

1598_book.fm Page 167 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

167

∆F = ( ∆F ′′)2 + ( ∆F ′)2 .

4.4.4

(4.92)

The Wide-Band Vertical-Coverage Directional Diagram: The Square Waveform Pulsed Searching Signal

Let us assume that the condition ∆p << ∆v is true and that we can neglect the dependence between the directional diagram gv(ψ* + c*z) and the parameter z within the limits of duration of the pulsed searching signal. Then, in the case of the square waveform pulsed searching signal with the duration τp , we can write ∞

Rγ (τ) =

∑

[

(

sin π ∆Fτ τ 1 −

| µτ − nTp | τp

π ∆Fτ τ

n= 0

)]

.

(4.93)

Comparing Equation (3.31) and Equation (4.93), one can see that Rγ(τ) is the aspect-angle normalized correlation function of the target return signal ﬂuctuations formed by covering the linearly frequency-modulated pulsed searching signals with the deviation ∆Fτ given by Equation (4.70) and duration τ ′p = τpµ–1. Unlike the normalized correlation function given by Equation (3.31), the one given by Equation (4.93), rigorously speaking, is not a periodic function despite this function possessing the properties of periodicity. The aspect-angle normalized correlation function given by Equation (4.93) is shown in Figure 4.11. This function consists of dissimilar narrow waves with the width equal to 2 τ ′p and the period Tp′. At n = 0, due to the condition ∆Fττp << 1, as a rule, we can write R γ (τ)

τ 0

2T ′p

3T p′

4T ′p

5T ′p

6T ′p

7T ′p

8T ′p

FIGURE 4.11 The aspect angle normalized correlation function of the target return signal ﬂuctuations with the square waveform pulsed searching signal.

Copyright 2005 by CRC Press

1598_book.fm Page 168 Monday, October 11, 2004 1:57 PM

168

Signal and Image Processing in Navigational Systems

Rγ ( τ) = 1 −

| τ| . τ ′p

(4.94)

The function Rγ(τ) is the aspect-angle normalized correlation function within the limits of a single period of the square waveform pulsed searching signal. At n ≠ 0, the shape of the waves is not triangular, and when the value of n is high, the difference between the shape of the waves and the triangular shape is greater. The duration of the waves is equal to 2 τ ′p independent of the value of n. Under the condition ∆Fττp = ∆Fτn Tp′ = 0.5, the waves have the shape of the cosine curve within the limits of the interval [–0.5π, 0.5π]. If ∆Fτ nTp′ < 0 is true, the waves have spiked tops. If ∆Fτn Tp′ > 0.5 is satisﬁed, the waves have crevasses. When ∆Fτn Tp′ = 1, the crevasse is maximal and reaches zero. When τ = n Tp′ , we can write Rγ ( τ) = sinc( π ∆Fτ τ) .

(4.95)

The function Rγ(τ) is the aspect-angle normalized correlation function of the interperiod ﬂuctuations in the glancing radar range. The normalized correlation function given by Equation (4.95) is shown in Figure 4.11 by the dotted line. The aspect-angle normalized correlation function Rγ(τ) given by Equation (4.95) is not an envelope of the normalized correlation function Rγ(τ) given by Equation (4.93). Therefore, the normalized correlation function Rγ(τ) is not equal to the product of the normalized correlation functions of intraperiod and interperiod ﬂuctuations because the shape of individual waves depends on the value of n. The normalized correlation function given by Equation (4.95) deﬁnes ﬂuctuations caused by difference in the radial velocities of scatterers with various aspect angles within the limits of the scanned element resolved in the radar range. The square waveform power spectral density with the effective bandwidth ∆Fτ given by Equation (4.70) at kp = 1 corresponds to the case considered here. In spite of the fact that with high values of the aspect angle, the effective bandwidth ∆Fτ can be high in value, for example, 200–600 Hz, the condition ∆Fττp << 1 is true, as a rule. For this reason, not only does the ﬁrst wave of the normalized correlation function given by Equation (4.93) have a triangular shape, but also a large number of the central waves has the shape closely resembling the triangular shape. Because the bandwidth ∆Fτ is less, the number of waves is high. Under the condition V → 0 or γ* → 0 the normalized correlation function Rγ(τ) given by Equation (4.93) tends to approach the following form ∞

Rγ (τ) =

∑ (1 − n= 0

Copyright 2005 by CRC Press

| µ τ p − nTp | τp

)

(4.96)

1598_book.fm Page 169 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

169

and the envelope of the power spectral density is determined by Equation (3.19). Consider the normalized correlation function of the interperiod ﬂuctuations in the ﬁxed radar range.40,41 If the condition τ = nTp [see Equation (4.93)] is satisﬁed, we can write sin[π ∆Fτ τ(1 − ∆F ′|τ|)] , π ∆Fτ τ

(4.97)

µ 2V = ⋅ cos β 0 cos γ ∗ . τp cτp

(4.98)

Rγ (τ) = where ∆F ′ =

Compare this with Equation (4.82). The normalized correlation function Rγ(τ) given by Equation (4.97) is not periodic. Comparing Equation (3.31) and Equation (4.97) under the condition µ = 1, one can see that Rγ(τ) is the aspectangle normalized correlation function of the ﬂuctuations formed by covering the square waveform linear-frequency-modulated pulsed signals with the duration τ ′c = ∆1F ′ and deviation ∆Fτ. These pulsed signals are formed by reﬂection from individual elementary scatterers moving relative to the moving radar and scanned with various values of the aspect angle within the limits of the element resolved in the radar range. Changes in the aspect angle during scatterer motion lead to changes in the Doppler frequency, i.e., to frequency modulation. The pulsed signal duration τ′ is the time required so that the scatterer with the coordinates β0 and γ* and moving with the velocity V could cover the scanned element resolved in the radar range. The character of the normalized correlation function Rγ(τ) of the interperiod ﬂuctuations given by Equation (4.97) depends on the ratio 2 2 3 ∆Fτ c τ p sin γ * . = ∆F ′ 2 hλ cos 2 γ *

(4.99)

Under the condition ∆F′ >> ∆Fτ , based on Equation (4.97), we can write Rγ (τ) = 1 − ∆F| ′ τ| .

(4.100)

This takes into account the ﬂuctuations caused by the exchange of scatterers within the limits of the scanned element resolved in the radar range with the moving radar (see Section 3.4). These ﬂuctuations are caused by the interperiod ﬂuctuations in the ﬁxed radar range. The power spectral density

Copyright 2005 by CRC Press

1598_book.fm Page 170 Monday, October 11, 2004 1:57 PM

170

Signal and Image Processing in Navigational Systems Sγ (ω ) ≅ sinc 2

( ) ω 2∆F′

(4.101)

corresponds to Rγ(τ) given by Equation (4.100). The effective bandwidth of the power spectral density Sγ(ω) given by Equation (4.101) is equal to ∆F′ [see Equation (4.98)]. Thus, we have the power spectral density of the square waveform target return signals with the duration τ ′c = ∆1F ′ . Under the condition ∆Fτ >> ∆F′, we can obtain Rγ(τ) given by Equation (4.95) based on Equation (4.97). Under this condition, the normalized correlation function in the glancing radar range coincides with that in the ﬁxed radar range because the value of the effective bandwidth ∆F′ is inﬁnitesimal. Comparing the effective bandwidths ∆Fτ and ∆F′ given by Equation (4.70) and Equation (4.98), respectively, we can conclude that at high values of the aspect angle, the Doppler ﬂuctuations play a main role, and with low values of the aspect angle, the ﬂuctuations caused by the exchange of scatterers ∆F play a main role. For arbitrary values of the ratio ∆Fτ′ , the power spectral density is given by Equation (3.32) if we replace the parameters ω ′M and τ ′p with the parameters ∆Ωτ and ∆1F ′ , respectively. Under this condition, both the determination of τc in Equation (3.36) and Figure 3.5–Figure 3.7 are true.

4.5 4.5.1

The Pulsed Searching Signal with the Moving Radar: The Azimuth Correlation Function General Statements

The azimuth-normalized correlation function is determined by Equation (4.7). For simplicity, we assume that the radar moves horizontally, i.e., ε0 = 0 and the vertical-coverage directional diagram is independent of the angle ψ*. In this case, the azimuth-normalized correlation function can be written in the following form:

∫

Rβ (τ) = N gh2 (ϕ) ⋅ e

ϕ  jτ Ωmax cos β 0 + cos γ * cos γ *  

dϕ .

(4.102)

The azimuth-normalized correlation function Rβ(τ) deﬁnes the ﬂuctuations caused by differences in the Doppler frequency of scatterers scanned with various values of the azimuth angle within the directional diagram in the horizontal plane, i.e., the horizontal-coverage directional diagram.42 After multiplying by the factor e jω 0τ and using the Fourier transform for Rβ(τ), we can write the power spectral density of the Doppler ﬂuctuations in the following form:

Copyright 2005 by CRC Press

1598_book.fm Page 171 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

Sβ (ω ) ≅

∫ g (ϕ) ⋅ δ[ω − ω 2 h

0

− Ω max cos(β 0 +

ϕ cos γ *

171

) cos γ * ] dϕ .

(4.103)

Let us introduce a new variable Ω = Ω max ⋅ cos(β 0 +

ϕ cos γ *

) cos γ * ,

(4.104)

where Ω is the Doppler shift in frequency for the scatterer with the coordinates β = β0 + cosϕγ * and γ = γ* . Using the ﬁltering property of the delta function,43,44 we can write the power spectral density of Doppler ﬂuctuations in the following form:

Sβ (ω ) ≅

gh2 [(arccos

ω − ω0 Ωmax cos γ *

1−

(

− β 0 ) cos γ * ]

ω − ω0 Ωmax cos γ *

)

2

+

gh2 [( − arccos 1−

ω − ω0 Ωmax cos γ *

(

− β 0 ) cos γ * ]

ω − ω0 Ωmax cos γ *

)

2

,

(4.105) where arccos (x) and β0 have the same sign. The power spectral density Sβ(ω) given by Equation (4.105) — the power per unit bandwidth — is the ratio between the power dp of the target return signal from the scatterers, which are skewed with respect to the direction of the moving radar under the angle ±β with the same Doppler frequency, dp ≅ [ g h2 (ϕ) + g h2 ( −ϕ − 2β 0 cos γ * )] dϕ .

(4.106)

In addition, the bandwidth dω occupied by the target return signal is determined by dω = d[ω 0 + Ω max cos(β 0 +

ϕ cos γ *

) cos γ * ] = − Ω max sin(β 0 + cosϕγ ) dϕ . *

(4.107) In this case, we can write Sβ(ω) in the following form: Sβ (ω ) ≅

g h2 (ϕ) + g h2 ( −ϕ − 2β 0 cos γ * ) sin(β 0 +

ϕ cos γ *

)

,

(4.108)

where ϕ and Ω are related by Equation (4.104). Based on Equation (4.104), we can write ω−ω

ϕ = (arccos Ωmax cos0 γ * − β 0 ) cos γ * .

Copyright 2005 by CRC Press

(4.109)

1598_book.fm Page 172 Monday, October 11, 2004 1:57 PM

172

Signal and Image Processing in Navigational Systems

Substituting Equation (4.109) in Equation (4.108), we can obtain the power spectral density of the Doppler ﬂuctuations given by Equation (4.105). This physical representation is the basis of the technique for determining the power spectral density of the Doppler ﬂuctuations. Reference to Equation (4.105) shows that we can obtain the following boundary values for frequency ω: ω min = ω 0 − Ω max cos γ * ≤ ω ≤ ω 0 + Ω max cos γ * = ω max

(4.110)

independently of the shape of the directional diagram. The power spectral density Sβ(ω) at frequencies ωmin and ωmax tends to approach ∞ if a coefﬁcient of ampliﬁcation of the radar antenna in the corresponding directions does not equal zero. The frequency-modulated searching signal has the same power spectral density under the slow harmonic frequency modulation within the limits of the interval [– Ωmax cos γ*, Ωmax cos γ*]. This circumstance does not contradict the physical representations and can be explained by the fact that the bandwidth |dΩ| = Ωmax sin β cos γ * dβ

(4.111)

of the target return signal within the limits of the azimuth angle dβ tends to approach zero as β → 0, but the target return signal power dp = 0.5cτp dβ

(4.112)

remains ﬁnite. The presence of two terms in the numerator in Equation (4.105) can be explained as follows. Scatterers disposed under the same angle at the left and right from the direction of the moving radar can generate the same Doppler frequency (see Figure 4.12). Consider, for example, the nondirected antenna in the horizontal plane radar antenna. Reference to Equation (4.105) shows that Sβ (ω ) ≅ 1−

(

2 ω − ω0 Ωmax cos γ *

)

2

.

(4.113)

The power spectral density Sβ(ω) given by Equation (4.113) is shown in Figure 4.13 by the dotted line. At the frequencies ωmin and ωmax , Sβ(ω), as was mentioned previously, tends to approach ∞. The analogous power spectral density in scanning the three-dimensional (space) target is shown in Figure 4.13 by the horizontal dotted line. In the case of the three-dimensional (space) target, the maximum and minimum Doppler frequencies of Sβ(ω) are determined by Equation (3.46) that is based on the use of Equation (4.110) under the condition γ* = 0. Sβ(ω) is ﬁnite at these frequencies. The corresponding Copyright 2005 by CRC Press

1598_book.fm Page 173 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

173

V

ϕ β0

− ϕ − 2 β0 cos γ

*

FIGURE 4.12 Formation of the power spectral density Sβ(ω).

azimuth-normalized correlation function Rβ(τ) is determined in the following form: Rβ ( τ) = J 0 (Ω max τ cos γ * ) ⋅ e jω 0τ ,

(4.114)

where J0(x) is the Bessel function of the ﬁrst order. Reference to Equation (4.105) shows that if the horizontal-coverage directional diagram is narrow, then a small part of the power spectral density Sβ(ω) of the Doppler ﬂuctuations given by Equation (4.113) can be cut. The shape of the cut power spectral density depends both on the form of the horizontal-coverage directional diagram and on the position of the radar antenna axis with respect to the direction of the moving radar. Deformation of Sβ(ω) for various positions of the pencil-beam radar antenna with respect to the direction of moving radar without consideration of the side lobes is shown in Figure 4.13. If the horizontal-coverage directional diagram is narrow, the power spectral density Sβ(ω) given by Equation (4.105) can be simpliﬁed in two important cases: (1) the radar antenna is high deﬂected from the direction of moving radar, and (2) it is low deﬂected. The simplest way to do this is to introduce an approximate factor in Equation (4.102):

Copyright 2005 by CRC Press

1598_book.fm Page 174 Monday, October 11, 2004 1:57 PM

174

Signal and Image Processing in Navigational Systems S β (ω) 9

1

2

8

3

7

6

5

4

ω − ω0 −Ωmax

0 Ωmax cos γ

Ωmax

−Ωmax cos γ

*

*

FIGURE 4.13 The power spectral density Sβ(ω) with the nondirected (dotted line) and pencil-beam (solid line) radar antenna at various values of β0: (1) β0 = 0°; (2) β0 = 10°; (3) β0 = 30°; (4) β0 = 60°; (5) β0 = 90°; (6) β0 = 120°; (7) β0 = 150°; (8) β0 = 170°; (9) β0 = 180°.

Ω(ϕ) ≅ Ω∗ − ϕ Ω max sin β 0 − ϕ 2 Ω max cos β 0 ,

(4.115)

Ω∗ = Ω max cos β 0 cos γ *

(4.116)

where

is the Doppler frequency corresponding to the middle of the scanned surface element of the two-dimensional target. Equation (4.116) is different from Equation (3.48) and Equation (3.49) by the exchange of the angle γ0 for the angle γ* . 4.5.2

The High-Deflected Radar Antenna

With the high-deﬂected radar antenna, if the value of sin β0 is not so low, we can neglect the third term in Equation (4.115). Then the azimuth-normalized correlation function Rβ(τ) of the target return signal Doppler ﬂuctuations can be written in the following form:

Copyright 2005 by CRC Press

1598_book.fm Page 175 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

∫

Rβ ( τ) = N ⋅ e jΩ* τ g h2 (ϕ) ⋅ e − jϕ Ωmax τ sin β0 dϕ .

175

(4.117)

Multiplying Equation (4.117) with the factor e jω 0τ , we can write44

(

Sβ (ω ) ≈ gh2 −

ω − ω 0 − Ω* Ωmax sin β 0

).

(4.118)

With the high-deﬂected radar antenna, the power spectral density Sβ(ω) coincides approximately in shape with the square of the directional diagram by power — the receiving and transmission directional diagrams — at the cross section through the scanned surface element of the two-dimensional target resolved in the radar range.45 The analogous conclusion was made by investigation of the three-dimensional (space) target, although the directional diagrams are different for these cases. This phenomenon can be easily explained based on the physical meaning. Reference to Equation (4.118) shows that the effective bandwidth of the power spectral density Sβ(ω) is related to the effective widths ∆* and ∆(*2 ) of the directional diagram by power at the cross section ψ* by the following relationships: ∆Fh = 2V λ ∆ (*2 ) sin β 0 = 2V λ kh ∆ * sin β 0 ,

(4.119)

where kh is the coefﬁcient of the shape of the directional diagram (see Section 2.4). Equation (4.119) is similar to Equation (3.76) in the case of the threedimensional (space) target. Using Equation (4.115) and Equation (4.118) for the Gaussian and sincdirectional diagrams, we can obtain the same formulae as in Section 3.2: the power spectral densities of the Doppler ﬂuctuations are determined by Equation (3.80) and Equation (3.86), respectively, and the normalized correlation functions are given by Equation (3.79) and Equation (3.88), respectively. The effective bandwidth of the power spectral density Sβ(ω) and the correlation interval are determined by Equation (4.119), where k h = 12 and 23 , respectively. Figure 3.12 and Figure 3.13 are true for the cases considered here. In particular, with the Gaussian directional diagram, we can write

Sβ (ω ) ≅ e

 ω − ω 0 − Ω*  −π ⋅  2 π ∆Fh 

2 2

Rβ (τ) = e − π ∆Fh τ

Copyright 2005 by CRC Press

2

;

+ j ( ω 0 + Ω* ) τ

(4.120)

.

(4.121)

1598_book.fm Page 176 Monday, October 11, 2004 1:57 PM

176

Signal and Image Processing in Navigational Systems

4.5.3

The Low-Deflected Radar Antenna

Let us consider the opposite case, i.e., the radar antenna is low deﬂected from the direction of the moving radar. When the value of the angle β0 is low and the beam width of the horizontal-coverage directional diagram is narrow, the power spectral density Sβ(ω) of the Doppler ﬂuctuations is close to the maximum Doppler frequency ω0 + Ωmax cos γ* . We can assume that ω − ω0 ≅ 1. Then we can write Ωmax cos γ ∗ arccos

ω − ω0 Ωmax cos γ *

≅

2 1 − 

ω −ω 0 Ωmax cos γ *

= 

2Ω Ωmax*

,

(4.122)

where Ω = ω 0 + Ω max cos γ * − ω

Ω max* = Ω max cos γ *

and

(4.123)

Reference to Equation (4.105) shows that at the low-deﬂected radar antenna, the power spectral density Sβ(ω) can be written in the following form: Sβ (ω ) ≅

Ωmax* Ω

  ⋅  gh2   

2Ω Ωmax*

  − β 0  cos γ *  + gh2  −   

2Ω Ωmax*

  − β 0   cos γ *  ,   (4.124)

where Ω ≥ 0, i.e., ω ≤ ω0 + Ωmax cos γ* If β = 0° and the horizontal-coverage directional diagram is symmetric, we can write Sβ (ω ) ≅

Ωmax* Ω

⋅ gh2  

cos γ *  . 

2Ω Ωmax*

(4.125)

The shape of the power spectral density Sβ(ω) at low values of the angle β0 is greatly different from the shape of the square of the directional diagram. It tends to approach ∞ at the maximum frequency. Sβ(ω) is asymmetric and is shown in Figure 4.14 with the Gaussian low-deﬂected directional diagram. In this case, we can write

Sβ (ω ) ≅

∆Ω0 – ∆ΩΩ ⋅e Ω

− η2 0

(

ch 2 η

Ω ∆Ω0

),

(4.126)

where ∆Ω 0 =

Copyright 2005 by CRC Press

Ω max* ∆(*2 ) 4 π cos 2 γ *

=

V ∆(*2 ) ⋅ cos γ * ; λ

(4.127)

1598_book.fm Page 177 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

177

S β (ω)

1

1.0 0.9 0.8 0.7 0.6 0.5

2

0.4 3 0.3 0.2

4

−16

−14

−12

−10

−8

−6

−4

ω0 + Ωmax cos γ − ω * ∆Ω0 −2

0

FIGURE 4.14 The power spectral density Sβ(ω) with the low-deﬂected radar antenna: (1) cos γ* = 0.5; (3)

β0 ∆∗

cos γ * =

1 2

; (4)

β0 ∆∗

β0 ∆∗

cos γ* = 0; (2)

β0 ∆∗

cos γ* = 1.

η = 2π ⋅

β0 ⋅ cos γ * . ∆*

(4.128)

At β = 0°, we can write Sβ (ω ) ≅

∆Ω 0 − ⋅e Ω

Ω ∆Ω 0

.

(4.129)

As Sβ(ω) tends to approach ∞ at the maximum frequency, it is impossible to deﬁne both the effective bandwidth and bandwidth of Sβ(ω) at any level with respect to the maximum. It is necessary to deﬁne the bandwidth of Sβ(ω) without normalization. Let us introduce, for example, the effective bandwidth ∆Fh , in which there is the same total power that is within the limits of the effective bandwidth of Sβ(ω), for example, 80%, with Gaussian and sinc-directional diagrams. With this deﬁnition of the effective bandwidth, we can write ∆Fh = if β 0 < β ′0 =

kh ∆ ∗ 2 cos γ ∗

k h2 V∆(*2 ) 2 β 0 cos γ * ⋅ 1+ kh ∆ * 4 λ cos γ *

(

)

(4.130)

, where kh is the same as in Equation (4.119). Under the

condition β 0 ≥ β ′0 , the deﬁnition of ∆Fh in Equation (4.119) is true and we Copyright 2005 by CRC Press

1598_book.fm Page 178 Monday, October 11, 2004 1:57 PM

178

Signal and Image Processing in Navigational Systems

obtain the same result as in Equation (4.130) under the condition β 0 = β ′0 . The shape of the power spectral density Sβ(ω), rigorously speaking, is very different from that given by Equation (4.118). Comparing Equation (4.119) and Equation (4.130), one can see that the effective bandwidth of the power spectral density Sβ(ω) of the Doppler ﬂuctuations at β0 = 90° is approximately ∆10∗ times that at β0 = 0°. The lower the width of the directional diagram in value, the greater the difference in the effective bandwidth of Sβ(ω). In the case of the narrow-band directional diagram, this difference is nearly a few hundreds. The effective bandwidth is not more than 2–3 Hz or even 0.3–0.5 Hz in speciﬁc cases. The same result is observed in scanning the three-dimensional (space) target.

4.6

4.6.1

The Pulsed Searching Signal with the Moving Radar: The Total Correlation Function and Power Spectral Density of the Target Return Signal Fluctuations General Statements

The total normalized correlation function R(t, τ) of the target return signal ﬂuctuations with the pulsed searching signal is deﬁned by the product of the aspect-angle normalized correlation function Rγ(t, τ) and the azimuthnormalized correlation function Rβ(τ): R(t , τ) = Rγ (t , τ) ⋅ Rβ ( τ) .

(4.131)

Rγ(t, τ), given by Equation (4.66), deﬁnes the rapid ﬂuctuations in the radar range and the slow interperiod ﬂuctuations caused by the difference in the radial components of the velocity of scatterers in the aspect-angle plane. The azimuth-normalized correlation function Rβ(τ) given by Equation (4.102) deﬁnes only the interperiod ﬂuctuations caused by the difference in the radial components of the velocity of scatterers in the azimuth plane. For this reason, the product of Rγ(t, τ) and Rβ(τ) leads to changes only in the interperiod ﬂuctuations and does not act on the ﬂuctuations in the radar range. The interperiod ﬂuctuations in the glancing radar range, i.e., when the condition τ = nTp′ is true, are deﬁned by the product of the aspect-angle normalized correlation function Rγ(t, τ) given by Equation (4.67) and the azimuth-normalized correlation function Rβ(τ) given by Equation (4.102). The total power spectral density in the glancing radar range is deﬁned by the convolution between the aspect-angle power spectral density Sγ[ω, ψ* (t)] given by Equation (4.68) and the azimuth power spectral density Sβ(ω) given by Equation (4.105). If the shape of the directional diagram in the vertical and horizontal planes is arbitrary, the shape of the pulsed searching signal Copyright 2005 by CRC Press

1598_book.fm Page 179 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

179

is also arbitrary, and there is a linear approximation for the variable Ω(ϕ) in Equation (4.115), and the total power spectral density of the interperiod ﬂuctuations is the convolution between Sγ[ω, ψ* (t)] given by Equation (4.68) and Sβ(ω) given by Equation (4.118): S(ω ) ≅

∫ Π (− 2

x c∗Ω γ

) ⋅ gh2 (ψ * − Ωx ) ⋅ g(− ω − ω Ω− Ω − x ) dx . 0

γ

*

h

(4.132)

In particular, with the square waveform pulsed searching signal, we can write 0.5 ∆Ωτ

S(ω ) ≅

∫

gh2 (ψ * −

x Ωγ

) ⋅ g(−

ω − ω 0 − Ω* − x Ωh

) dx .

(4.133)

− 0.5 ∆Ωτ

The interperiod ﬂuctuations in the ﬁxed radar range are deﬁned by the product of the aspect-angle normalized correlation function Rγ(t, τ) given by Equation (4.66) at the condition τ = nTp and the azimuth-normalized correlation function Rβ(τ) given by Equation (4.102). In the general case, we must apply various numerical techniques to deﬁne the power spectral densities Sβ(ω) and Sγ[ω, ψ* (t)].46 Let us consider two particular cases in more detail at the Gaussian directional diagram: (1) the Gaussian pulsed searching signal and (2) the square waveform pulsed searching signal. During consideration of these particular cases, we do not use numerical techniques.

4.6.2

The Gaussian Directional Diagram: The Gaussian Pulsed Searching Signal

With the Gaussian directional diagram and when the pulsed searching signal is Gaussian, we can write the total normalized correlation function R(t, τ) of the target return signal ﬂuctuations in the following form [see Equations (4.6), (4.73), (4.74), (4.84), and (4.121)]: ∞

R[τ , ψ (t)] = Rγ′ (τ) ⋅ Rγ′′(τ , ψ * ) ⋅ Rβ (τ) ⋅ e

jω 0τ

=e

− π ∆F 2 τ 2 + j ( ω 0 + ΩD ) τ

⋅

∑e

 τ − nTp  − π ⋅ 2  τ ′p 

2

,

n= 0

(4.134) where 2  ∆( )  ΩD = Ω* + Ω′′ = Ωmax cos β 0 cos γ * + ψ * Ω γ  1 + (v2)  ; ∆p  

Copyright 2005 by CRC Press

(4.135)

1598_book.fm Page 180 Monday, October 11, 2004 1:57 PM

180

Signal and Image Processing in Navigational Systems

∆F 2 = ∆Fh2 +

∆Fτ2 1+

∆(p2 )

.

(4.136)

∆(v2 )

The power spectral density of the ﬂuctuations corresponding to the total normalized correlation function R[τ, ψ (t)], which is determined by Equation (4.134), is analogous to the spectral power density Sγ(ω, ψ*) given by Equation (4.90): ∞

S(ω , ψ * ) ≅

∑e

 nΩ′p  −π ⋅  ∆Ω′p 

2

⋅e

 ω − ΩD − nΩ′p  −π ⋅ ∆Ω  

2

,

(4.137)

n= 0

where ∆Ω′p is given by Equation (4.79). As one can see from Equation (4.137), the power spectral density S(ω, ψ*) of the ﬂuctuations is different from the power spectral density Sγ(ω, ψ*) given by Equation (4.90) by the width of the waves ∆Ω = 2π∆F, where the effective bandwidth ∆F is determined by Equation (4.136) instead of Equation (4.86), and by their position on the frequency axis; the shift with respect to the value nΩ ′p is given by ΩD = Ω* + Ω″, where ΩD is given by Equation (4.135), instead of the value Ω″ given by Equation (4.85). The envelope of comb waves is the same as for Sγ(ω, ψ*) given by Equation (4.90). The frequency ΩD characterizes the Doppler shift in the frequency of the power spectral density of the slow ﬂuctuations. The value of ΩD depends on the ratio

∆v ∆p

. With the narrow-band pulsed searching signal, when the

condition ∆p << ∆v is true, we can neglect the second term in Equation (4.135), i.e., ΩD = Ω*. This means that the average frequency of the power spectral density S(ω, ψ*) is deﬁned by the direction to the center of the scanned element resolved in the radar range, i.e., cos β0 cos γ* = cos θ*. In doing so, the vertical-coverage directional diagram does not act on the frequency ΩD. In the case of the broadband pulsed searching signal, if the condition ∆p >> ∆v is true, we can write Ω D → Ω 0 = Ω max cos β 0 cos γ 0 ,

(4.138)

i.e., it is deﬁned by the direction of the directional diagram axis: cos β0 cos γ0 = cos θ0. In other cases, some intermediate value of the angle γ, which is within the limits of the interval [γ*, γ0], is the deﬁnitive one. Consider the interperiod ﬂuctuations, the power spectral density of which nT

is transformed as a result of the convolution. Under the condition τ = µp , based on Equation (4.134), we can write the normalized correlation function

Copyright 2005 by CRC Press

1598_book.fm Page 181 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

181

of the interperiod ﬂuctuations in the glancing radar range in the following form: R(τ , ψ * ) = e − π ∆F

2 2

τ + j ( ω 0 + ΩD ) τ

.

(4.139)

The corresponding power spectral density takes the following form: ∞

S(ω , ψ * ) ≅

∑e

ω − ΩD  − π ⋅   ∆Ω 

2

.

(4.140)

n= 0

The normalized correlation function R(τ, ψ*) given by Equation (4.139) is the envelope of the comb waves of the normalized correlation function R[τ, ψ(t)] given by Equation (4.134). The power spectral density S(ω, ψ*) given by Equation (4.140) is the central wave of the power spectral density S(ω, ψ*) given by Equation (4.137). The effective bandwidth of S(ω, ψ*) given by Equation (4.140) depends on the ratio

∆p ∆v

. Under the condition ∆p << ∆v , we can write this in the following

form: ∆F = ∆Fh2 + ∆Fτ2 = 2 Vλ−1 ∆(h2 ) sin 2 β 0 + ∆(p2 ) cos 2 β 0 sin 2 γ * . (4.141) Under the condition ∆p << ∆v , the effective bandwidth can be written in the following form also: 2 (2) ∆F = ∆Fh2 + ∆Fv2 = 2 V λ −1 ∆ h sin 2 β 0 + ∆ (v ) cos 2 β 0 siin 2 γ 0 . (4.142)

The effective bandwidth ∆F of S(ω, ψ*) given by Equation (4.142) is the same as in the case of scanning the three-dimensional (space) target [see Equation (3.81)]. In particular, when the directional diagram is axial symmetric, i.e., when the condition ∆h = ∆p = ∆a is satisﬁed, we can write this in the following form: ∆F = 2 V λ −1 ⋅ ∆ a sin θ0 ,

(4.143)

where θ0 is the angle between the directional diagram axis and the vector of velocity of the moving radar. Equation (4.143) is similar to Equation (3.76). Equation (4.141) and Equation (4.142) are equivalent. Reference to Equation (4.141) shows that the angle width ∆p of the surface element of the twodimensional target resolved in the radar range plays a role of the verticalcoverage directional diagram width ∆v . The ﬁrst term both in Equation Copyright 2005 by CRC Press

1598_book.fm Page 182 Monday, October 11, 2004 1:57 PM

182

Signal and Image Processing in Navigational Systems

I

β0 II

II

ρ

I

FIGURE 4.15 Regions of the interperiod target return signal ﬂuctuations in the glancing radar range.

(4.141) and Equation (4.142) is deﬁned by the difference in the Doppler frequencies — the radial components of velocity of the moving scatterers — in the horizontal plane within the horizontal-coverage directional diagram. The second term, both in Equation (4.141) and Equation (4.142), is deﬁned by the difference in the Doppler frequencies in the vertical plane within the limits of the scanned surface element of the two-dimensional target resolved in the radar range [see Equation (4.141)] or of the vertical-coverage directional diagram [see Equation (4.142)]. The contribution of the ﬁrst and second terms [see Equation (4.141) and Equation (4.142)] in the effective bandwidth ∆F of the power spectral density S(ω, ψ*) of the interperiod ﬂuctuations is shown in Figure 4.15, where bounds of “equal contributions” are on the plane (β0, ρ). These bounds are determined using the following formula: h −1ρ = cosec γ * = k ctg β 0 , where k =

∆p ∆h

for Equation (4.162) and k =

∆v ∆h

(4.144)

for Equation (4.163). In the latter

case, we must use the angle γ0 instead of the angle γ* in Equation (4.144). Near the boundaries, the target return signal ﬂuctuations are caused by the difference in the Doppler frequencies both in the azimuth plane and in the

Copyright 2005 by CRC Press

1598_book.fm Page 183 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

183

aspect-angle plane. Under the ﬁxed azimuth angle, with an increase in the radar range, the difference in the radial velocities in the azimuth plane is increased, and with a decrease in the radar range, the difference in the radial velocities in the aspect-angle plane is increased. The scanned surface of the two-dimensional target is divided on two regions: I and II. The main source of the ﬂuctuations in region I is a difference in the radial velocities in the azimuth plane, whereas that in region II is a difference in the radial velocities in the aspect-angle plane. In the boundary region, it is necessary to take into consideration the two sources of ﬂuctuations [see Equation (4.136) and Figure 4.15]. Under the condition τ = nTp , based on Equation (4.134), we can write the normalized correlation function of the interperiod ﬂuctuations in the ﬁxed radar range in the following form: R(τ , ψ * ) = e − π [∆F

2

+ ( ∆F′ )2 ]τ 2 + j ( ω 0 + ΩD ) τ

.

(4.145)

The normalized correlation function R(τ, ψ*) in Equation (4.145) is the product of the aspect-angle normalized correlation function Rγ′ ( τ) given by Equation (4.80), the aspect-angle normalized correlation function Rγ′′ (τ, ψ*) given by Equation (4.84), and the azimuth-normalized correlation function Rβ(τ) given by Equation (4.121). The power spectral density in the ﬁxed radar range corresponding to R(τ, ψ*) in Equation (4.145) is equivalent to the power spectral density S(ω, ψ*) determined by Equation (4.140) if we change the effective bandwidth ∆Ω for the effective bandwidth ∆Ω 2 + ( ∆Ω ′)2 , where ∆Ω′ is determined by Equation (4.82). The component ∆Ω′ takes into consideration the ﬂuctuations caused by the exchange of scatterers and is very low in value, as a rule. For this reason, the value of ∆Ω′ is taken into account if the effective bandwidth is low in value, i.e., with low values of the angles β0 and γ* or γ0. In this case, we have three regions (see Figure 4.16). Regions I and II are the same as in Figure 4.15. Region III deﬁnes the ﬂuctuations caused by the exchange of scatterers in the scanned surface element of the two-dimensional target resolved in the radar range. In more rigorous investigation, we should consider the following: as β → 0, the effective bandwidth ∆Fh of the power spectral density S(ω, ψ*) of the interperiod ﬂuctuations in the ﬁxed radar range, does not tend to approach zero and remains a ﬁnite value, but it is low [see Equation (4.130)]. Because of this, the ﬂuctuations caused by the exchange of scatterers and those caused by the difference in the Doppler frequencies in the azimuth plane have approximately the same contribution. For example, at β0 = 0°, ∆h = 2°, τp = 1 µsec, λ = 3 cm, and if the angle γ* is not so high in value, the effective bandwidth is deﬁned by the form ∆Fh ≅ ∆F ′.

Copyright 2005 by CRC Press

1598_book.fm Page 184 Monday, October 11, 2004 1:57 PM

184

Signal and Image Processing in Navigational Systems

I

β0 III

II

II

ρ

III

I

FIGURE 4.16 Regions of the interperiod target return signal ﬂuctuations in the ﬁxed radar range.

4.6.3

The Gaussian Directional Diagram: The Square Waveform Pulsed Searching Signal

Let us consider the total instantaneous normalized correlation function with the square waveform pulsed searching signal and the Gaussian directional diagram. Let us assume that the vertical-coverage directional diagram is wide, i.e., the condition ∆p << ∆v is satisﬁed and the radar antenna is high deﬂected. Otherwise, if the condition ∆p >> ∆v is satisﬁed, then, as was discussed in Section 4.3.4, the intraperiod ﬂuctuations are absent in the central part of the pulsed target return signal. They are present only at the leading and trailing edges of the pulsed target return signal.47 Then the total instantaneous normalized correlation function deﬁnes the intraperiod and interperiod target return signal ﬂuctuations. Based on Equation (4.93) and Equation (4.121), we can write this in the following form: ∞

R(τ , ψ * ) = Rβ (τ) ⋅ Rγ (τ , ψ * ) = e

− π ∆Fh2 τ 2 + j ( ω 0 + Ω* ) τ

⋅

∑ n= 0

(

| τ − nTp′ |

sin π ∆Fτ τ 1−

π ∆Fτ τ

τ ′p

)

, (4.146)

where the effective bandwidths ∆Fh and ∆Fτ are determined by Equation (4.70) and Equation (4.119), respectively. The total instantaneous normalized correlation function R(τ, ψ*) of the ﬂuctuations given by Equation (4.146) is shown in Figure 4.17 without the factor e j(ω 0 + Ω* ) τ . Copyright 2005 by CRC Press

1598_book.fm Page 185 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

185

R(τ, ψ ) * 3 2 1

τ 0

2T p′

3T ′p

4T ′p

5T ′p

6T ′p

7T ′p

8T ′p

FIGURE 4.17 The total instantaneous correlation function of the intraperiod and interperiod target return signal ﬂuctuations with the square waveform pulsed searching signal and Gaussian horizontalcoverage directional diagram of radar antenna — Rγ(τ, ψ*) is shown by the solid line and Rβ(τ) is shown by the dotted line: (1) ∆Fh >> ∆Fτ ; (2) ∆Fh ≈ ∆Fτ ; (3) ∆Fh << ∆Fτ.

To deﬁne the spectral power density of the ﬂuctuations corresponding to the total instantaneous normalized correlation function given by Equation (4.146) is very difﬁcult in the general case. However, under the condition ∆Fh >> ∆Fτ , reference to Figure 4.17 (curve 1) shows that the condition ∆Fτ τ ≤

∆Fτ << 1 ∆Fh

(4.147)

is true within the limits of the correlation interval of the azimuth-normalized correlation function Rβ(τ), i.e., if the condition τ ≤ ∆F1h is satisﬁed. If the condition given by Equation (4.147) is satisﬁed, the approximation made in Equation (4.96) is true for the aspect-angle normalized correlation function Rγ(τ). In terms of Figure 4.18, the total instantaneous normalized correlation function of the ﬂuctuations can be written in the following form: ∞

2 2

R(τ , ψ * ) = e − π ∆Fh τ

+ j ( ω 0 + Ω* ) τ

⋅

∑ (1 −

| τ − nTp′ |

n= 0

τ′p

).

(4.148)

The condition given by Equation (4.147) means that c τ p sin 3 γ * ∆Fτ = << 1 ∆Fh 2 h∆ * tg β 0 cos γ *

(4.149)

[see Equation (4.70) and Equation (4.119)]. Usually, this condition is satisﬁed when the directional diagram is high deﬂected from the direction of the

Copyright 2005 by CRC Press

1598_book.fm Page 186 Monday, October 11, 2004 1:57 PM

186

Signal and Image Processing in Navigational Systems R (τ, ψ*)

τ 2T p′

0

3T p′

4T ′p

5T ′p

FIGURE 4.18 The total instantaneous correlation function of the intraperiod and interperiod target return signal ﬂuctuations with the square waveform pulsed searching signal and Gaussian horizontalcoverage directional diagram of radar antenna — Rγ(τ, ψ*) is shown by the solid line and Rβ(τ) is shown by the dotted line: ∆Fh >> ∆Fτ .

moving radar in the horizontal plane and the aspect angles are not so high in value within the vertical-coverage directional diagram. The power spectral density of the ﬂuctuations corresponding to the normalized correlation function R(τ, ψ*) given by Equation (4.148) has the comb shape and consists of the partial Gaussian power spectral densities — the Gaussian waves — with the effective bandwidth ∆Fh and the envelope determined in the following form: ∞

S(ω ) ≅ sinc 2 [0.5(ω − ω 0 − Ω∗ ) τ p ]

∑e

 ω − ω 0 − Ω∗ − n Ω′p  −π ⋅  ∆Ωh2  

2

.

(4.150)

n= − ∞

For the considered approximation, the average Doppler frequency Ω* depends on the function ψ*(t) and is within the limits of the interval Ω∗ ∈[Ω max cos β 0 cos( γ 0 − ∆ v ), Ω max cos β 0 cos( γ 0 + ∆ v )] .

(4.151)

Let us consider the interperiod ﬂuctuations in the glancing radar range. With the two-dimensional Gaussian directional diagram, and based on Equation (4.133), we can write the power spectral density in the following form: 0.5 ∆Ωτ

S(ω ) ≅

∫

e

 ω−ω −Ω −x 2  ψ Ω −x2 ∗ γ   ∗ 0  − 2 π ⋅  + ∆Ωh   ∆Ωv     

dx .

(4.152)

− 0.5 ∆Ωτ

The integral in Equation (4.152) can be written using the error integrals. The particular cases can be obtained based on Equation (4.152). Copyright 2005 by CRC Press

1598_book.fm Page 187 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

187

If the duration of the pulsed searching signal is long, i.e., the condition ∆p >> ∆v is true, the limits of the integral in Equation (4.152) can be considered as inﬁnite. In terms of Ω0 = Ω* + ψ*Ωγ , we can write

S(ω ) ≅ e

 ω − ω 0 − Ω0  − 2π ⋅    ∆Ω2h + ∆Ω2v 

.

(4.153)

This power spectral density coincides with the power spectral density Sg(ω) of the slow ﬂuctuations given by Equation (4.28) with the simple harmonic searching signal at the conditions S°(γ) ≅ S°(γ0) and sin γ ≅ sin γ0. If the verticalcoverage directional diagram is wide, i.e., the condition ∆p << ∆v is satisﬁed, we can neglect the second term in Equation (4.152). Then, the power spectral density of the ﬂuctuations can be written in the following form: S(ω ) ≅ Φ( 2 π ⋅

ω − ω 0 − Ω* + 0.5 ∆Ωτ ∆Ωh

) − Φ(

2π ⋅

ω − ω 0 − Ω* − 0..5∆Ωτ ∆Ωh

).

(4.154)

This power spectral density is the result of the convolution of the Gaussian power spectral density Sβ(ω) given by Equation (4.120) and the square waveform power spectral density with the effective bandwidth ∆Ωτ given by Equation (4.70). In the arbitrary relationships between the values of ∆Ωτ and ∆Ωh , the power spectral density S(ω) given by Equation (4.154) has the shape of the “smoothed trapezium” placed symmetrically with respect to the frequency ω = ω0 + Ω* . The basis of the “smoothed trapezium” is equal to ∆Ωτ . The slope of the leading and trailing edges of the “smoothed trapezium” is equal to ∆Ωh. Under the condition ∆Ωτ ≥ 2∆Ωh , the bandwidth of S(ω) given by Equation (4.154) at the level 0.5 — the level at which the bandwidth of S(ω) is equal to the effective bandwidth — is equal to ∆Ωτ . The width of the leading edge and trailing edge between the levels 0.8 and 0.92 is equal to ∆Ω τ . Under the condition ∆Ωτ ≤ ∆Ωh , the power spectral density of the ﬂuctuations becomes Gaussian: 2 π

S(ω ) ≅ e

 ω − ω 0 − Ω*  − 2π ⋅   ∆Ωh 

2

.

(4.155)

The effective bandwidth ∆Ωh and average frequency ω0 + Ω* of the Gaussian power spectral density S(ω) given by Equation (4.155) are different from the effective bandwidth and average frequency of the power spectral density S(ω) given by Equation (4.153). The regions within the bounds of which one of the conditions — ∆Ωτ ≤ ∆Ωh or ∆Ωτ ≥ ∆Ωh — is satisﬁed are shown in Figure 4.15. These regions are the same as for the power spectral density S(ω, ψ*) given by Equation (4.140). If the directional diagram is not deﬂected, i.e., the condition β0 = 0° is true, the total power spectral density of ﬂuctuations is deﬁned by the convolution of the square waveform power spectral

Copyright 2005 by CRC Press

1598_book.fm Page 188 Monday, October 11, 2004 1:57 PM

188

Signal and Image Processing in Navigational Systems

density with the effective bandwidth ∆Ωτ given by Equation (4.70) and the power spectral density Sβ(ω) given by Equation (4.129). 4.6.4

The Pulsed Searching Signal with Low Pulse Period-to-Pulse Duration Ratio

The discussion up until now has been based upon the assumption that the pulse period-to-pulse duration ratio of the pulsed searching signal is large. Under this condition, not more than one scanned area of the Earth’s surface or the surface of the two-dimensional target is simultaneously within the vertical-coverage directional diagram at any ﬁxed instant of time. If the pulse period-to-pulse duration ratio is low in value and the duration of the pulsed searching signal is short, some simultaneously scanned areas of the Earth’s surface or the surface of the two-dimensional target are within the verticalcoverage directional diagram at each instant of time (see Figure 4.19). Their number is determined by f ≅

2 Tr Tp

. Because the target return signals from

these areas of the Earth’s surface are independent, then the correlation function of the resulting target return signal ﬂuctuations can be written in the following form: f

RΣen (t , τ) =

∑ i=1

f

R en (ti , τ) =

∑R

en

(t + iTp , τ) ,

(4.156)

i=1

where Ren(ti , τ) is the correlation function of the target return signal from the i-th scanned area of the Earth’s surface or the surface of the two-dimensional

FIGURE 4.19 Simultaneously scanned areas by the pulsed searching signal with low pulse period-to-pulse duration ratio.

Copyright 2005 by CRC Press

1598_book.fm Page 189 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

189

S Σ (ω, t )

ω FIGURE 4.20 The power spectral density S∑(ω, t) of the resulting target return signal at simultaneous scanning of areas by the pulsed searching signal with the low pulse period-to-pulse duration ratio.

target [see Equation (2.133)]. The power spectral density with the comb structure and with different values of p(ti) and Ω*(ti), in the general case, with different widths of waves, corresponds to each term of the sum in Equation (4.156). The power spectral density of the resulting target return signal is shown in Figure 4.20.

4.7

Short-Range Area of the Radar Antenna

Let us consider the ﬂuctuations in the short-range area of the radar antenna. In this case, it is necessary to account for the changes in values of the tracking angles ∆ϕrm and ∆ψrm of scatterers with the moving radar. We introduce a new variable z given by Equation (4.36). In this case, reference to Equation (2.122)–Equation (2.126) shows that the normalized correlation function of the ﬂuctuations in the short-range area of the radar antenna can be written in the following form: ∞

Rsren (t , τ) = N

∑ ∫∫ P[z − 0.5(µτ − nT )] ⋅ P [z + 0.5(µτ − nT )] *

p

p

n= 0

( × g ( ϕ + 0.5 ∆ϕ

× g ϕ − 0.5 ∆ϕ rm , ψ * + c* z − 0.5 ∆ψ rm rm

, ψ * + c* z + 0.5 ∆ψ rm

× e − jΩ(ϕ , ψ * + c* z ) τ dϕ dz ,

Copyright 2005 by CRC Press

) )

(4.157)

1598_book.fm Page 190 Monday, October 11, 2004 1:57 PM

190 where ∆ϕrm =

Signal and Image Processing in Navigational Systems Vτ ρ*

sin β0 and ∆ψrm =

Vτ ρ*

cos β0 sin γ*. For simplicity, we consider

the radar moving horizontally, i.e., ε0 = 0. Let the directional diagram be Gaussian. Then, based on Equation (4.157), we can write the normalized correlation function of the ﬂuctuations in the short-range area of the radar antenna in the following form: Rsren (t , τ) = R1 ( τ) ⋅ R2 (t , τ) ,

(4.158)

where R1 (τ) = e

∆F12 =

 ∆ϕ 2 ∆ψ 2rm  − 0.5π τ 2  rm +   ∆( 2 ) ∆(v2 )  h

2 2

= e − π ∆F1 τ ;

V 2  sin 2 β 0 cos 2 β 0 sin 2 γ *  ⋅ +  . 2ρ*  ∆ (h2 ) ∆ (v2 ) 

(4.159)

(4.160)

Here R1(τ) is the normalized correlation function of the ﬂuctuations caused by changes in the value of the aspect angle of scatterers. The function R2(t, τ) is the normalized correlation function in the long-range area where the directional diagram g(ϕ, ψ) is Gaussian. Naturally, R1(τ) deﬁnes the interperiod ﬂuctuations and not the intraperiod ﬂuctuations. The results discussed in Section 4.6, in which we consider the ﬂuctuations in the long-range area when the directional diagram g(ϕ, ψ) is Gaussian and the pulsed searching signal has the Gaussian and square waveform shapes, are true for the normalized correlation function R2(t, τ). We can use the results discussed in Section 4.6.2 — the pulsed searching signal is Gaussian — for the normalized correlation function R1(τ) given by Equation (4.159). In this case, we have to consider the following fact. Taking into consideration changes in the values of the angles of tracking ∆ϕrm and ∆ψrm , we should use the effective power spectral density bandwidth ~ ∆F of the ﬂuctuations instead of the effective bandwidth ∆F used in Section ~ 4.6.2. ∆F can be determined in the following form: ∆F˜ = ∆F 2 + ∆F12 ,

(4.161)

where the effective bandwidth ∆F1 is given by Equation (4.160) and the effective bandwidth ∆F is given by Equations (4.136), (4.141), and (4.142). Under the condition ∆h = ∆v = ∆a , ∆F1 can be written in the form ∆F1 =

V 2 ρ* ∆ a

× sin θ*, where θ* is the angle between the vector of velocity of the moving radar and the direction (β0, γ*) in which the radar is moving. Thus, for the case considered here, the normalized correlation function and power spectral density of the interperiod ﬂuctuations are deﬁned by the Copyright 2005 by CRC Press

1598_book.fm Page 191 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

191

Gaussian law, as in the case of the long-range area. The distinction is only in the case of the effective bandwidth ∆F1, which is different from the effective bandwidth ∆F given by Equation (4.136) by the additional term [see Equation (4.160)] decreasing with an increase in the radar range ρ. This additional term takes into consideration changes in the amplitude of elementary signals when the scatterers are covered by the directional diagram, and is deﬁned by the time of scanning the scatterers by the radar antenna beam — the time when the scatterers are within the limits of the radar antenna beam width. The value of ∆F1, unlike ∆F, is inversely proportional to the directional diagram width or the radar antenna beam width. Let us consider the particular cases. If the duration of the pulsed searching signal is short, i.e., the condition ∆p << ∆v is satisﬁed, the effective bandwidth ~ ∆F must be given by Equation (4.141), and the effective bandwidth ∆F can be determined in the following form:

∆F = V ⋅

 2 ∆ (p2 )  2 ∆ (h2 ) 1  2 1  + + sin β  2 + 2 ( 2 )  cos 2 β 0 sin 2 γ * .  2 0 2 (2)  2ρ* ∆ v  2ρ* ∆ h   λ  λ (4.162)

Reference to Equation (4.162) shows that, if the radar range is determined by ρ* = ρ′* =

λ 2 ∆(h2 )

, the ﬁrst term in Equation (4.162) is doubled. If the radar

range is given by ρ* = ρ′′* =

λ 2∆ h ∆ v

, the second term in Equation (4.163) is dou-

bled. The distance ρ′* deﬁnes the conditional boundary of the short-range area of the horizontal-coverage directional diagram, as follows from the theory of antennas. The distance ρ′′* can be determined in the following form: ρ*′′ =

∆ λ ⋅ v . (2) 2∆v ∆ p

Reference to Equation (4.163) shows that the distance ρ′′* is

(4.163) ∆v ∆p

>> 1 times

more than the distance to the boundary of the short-range area of the horizontal-coverage directional diagram. If the condition ∆p >> ∆v is satisﬁed or the searching signal is a continuous process, and therefore the conditions γ* = γ0 and ρ* = ρ0 are true, the effective bandwidth must be given by Equation (4.142). Then the effective bandwidth ~ ∆F can be determined in the following form: ∆F = V ⋅

Copyright 2005 by CRC Press

 2 ∆ (h2 )  2 ∆ (v2 ) 1  1  2 2 2  2 + 2 ( 2 )  sin β 0 +  2 + 2 ( 2 )  cos β 0 sin γ * . 2ρ0 ∆ v  2ρ0 ∆ h   λ  λ (4.164)

1598_book.fm Page 192 Monday, October 11, 2004 1:57 PM

192

Signal and Image Processing in Navigational Systems

The ﬁrst term in Equation (4.164) is the same as in Equation (4.162). The second term in Equation (4.164) is doubled when the radar range ρ0 is determined in the form ρ0 =

λ2 ∆(h2 )

, i.e., at the boundary of the short-range area

of the vertical-coverage directional diagram. Under the condition ∆h = ∆v = ~

∆a , the effective bandwidth ∆F can be determined in the following form: ∆F˜ =

ρ2 ρ2 2 V∆ a ⋅ 1 + sr2 ⋅ sin θ 0 = 1 + sr2 ⋅ ∆F , λ ρ0 ρ0

(4.165)

where ρ sr =

da2 λ = 2 ∆(a2 ) 2λ

(4.166)

is the conditional boundary of the short-range area, da is the diameter of radar antenna, and ∆F is the effective bandwidth given by Equation (4.143) in the long-range area. To assume the value of ρsr , consider the following instance: at da = 3 m, λ = 3 cm, we obtain the conditional boundary of the short-range area ρsr = 150 m, or at da = 0.01 m, λ = 5·10–7 m — in the case of the laser antenna — we obtain the conditional boundary of the short-range area ρsr = 100 m. Note that within the limits of the short-range area we must be very careful in using the previously mentioned formulae because if the radar antenna is not focused, there is no time for its directional diagram to form. For this reason, rigorously speaking, the previously mentioned formulae are true only in the case of the focused radar antennas. The power spectral density of the ﬂuctuations in the short-range area with the continuous searching signal is determined48 on the condition that the transmitting and receiving antennas are separated. The transmitting and receiving antennas are diverse during the distance 0 along the direction of the vector of velocity of the moving radar. The directional diagrams are the Gaussian axial symmetric directional diagrams with the same beam width. The axes of these directional diagrams are parallel, and being so, the target return signal is decreased due to the incomplete overlapping between the directional diagrams of the transmitting and receiving antennas, but the power spectral density of the ﬂuctuations is Gaussian and has the effective bandwidth given in Equation (4.165), as when the transmitting and receiving antennas are matched. The shift in the average frequency is determined in the following form: Ω ′0 = Ω 0 (1 − δ 0 ) ,

Copyright 2005 by CRC Press

(4.167)

1598_book.fm Page 193 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

193

where Ω0 =

4πV ⋅ cos β 0 cos γ 0 λ

and

δ0 =

02 ⋅ sin 2 θ 0 sin 2 γ 0 . (4.168) 2 h2

Thus, there is the additional shift δ0 in frequency depending on the ratio For example, at γ0 = 60°, θ0 = 70°,

0 h

0 h

.

= 0.25, we obtain the additional shift δ0

= 0.5%. With an increase in the value of the ratio h0 , the additional shift δ0 is sharply decreased. If the transmitting and receiving antennas are separated in the direction that is perpendicular to the vector of velocity of the moving radar, the additional shift δ0 is absent. We have to note that even if the transmitting and receiving antennas are matched, the radar moving during the time of propagation of the pulsed searching signal leads to incomplete overlapping of tracks of the directional diagrams of the transmitting and receiving antennas on the Earth’s surface or the surface of the two-dimensional target, making the target return signal more weak and giving rise to the shift in the average Doppler frequency. In the radar, these differences and values are inﬁnitesimal, but in the sonar, these differences and values are considerable.49

4.8 4.8.1

Vertical Scanning of the Two-Dimensional (Surface) Target The Intraperiod Fluctuations in Stationary Radar

In the vertical scanning of the two-dimensional (surface) target, the correlation function of the target return signal ﬂuctuations is determined by Equation (2.182). Let us assume that the radar is stationary and consider the intraperiod target return signal ﬂuctuations. The interperiod target return signal ﬂuctuations are absent. Therefore, the exponential factor in Equation (2.182) transforms into 1 and the correlation function of the target return signal ﬂuctuations becomes a periodic function with respect to the variable τ. For this reason, it is sufﬁcient to consider a single term (n = 0): 2 π θ2

R (t , τ ) = p 0 en 0

∫ ∫ g (θ cos α − ϕ , 2

0

0

θ sin α − ψ 0 ) ⋅ S (θ) sin θ dθ dα ,

θ1

(4.169)

Copyright 2005 by CRC Press

1598_book.fm Page 194 Monday, October 11, 2004 1:57 PM

194

Signal and Image Processing in Navigational Systems

where the limits of integration with respect to the variable θ depend on the variables t = Td + ∆t and τ: if

∆ t ≥ 0.5τ p ,

then

θ ∈[θ 1 , θ 2 ] at 0 ≤ 0.5|τ|≤ 0.5τ p ; (4.170)

 [0, θ 2 ] at 0 ≤ 0.5|τ|≤ ∆ t ; if 0 ≤ ∆ t ≤ 0.5τ p , then θ ∈   [θ1 , θ 2 ] at ∆ t ≤ 0.5|τ|≤ 0.5τ p ; (4.171) if − 0.5τ p ≤ ∆ t ≤ 0, then

θ ∈[0, θ 2 ] at 0 ≤ 0.5|τ|≤ ∆ t + 0.5τ p ; (4.172)

θ1,2 = θ∗2 ∓ c ⋅ ∆ t = t − Td

τ p −| τ | 2h

= 2 ∆ t∓

and

τ p −| τ | Td

;

(4.173)

Td = 2 hc −1 .

(4.174)

Let us limit the case of the axial symmetric nondeﬂected Gaussian directional diagram. Let us suppose that the speciﬁc effective scattering area S°(θ) is given by Equation (2.187). Then, in the region where the condition ∆t ≥ 0.5τp is satisﬁed [see Equation (4.170)], we can write θ2

∫

R (t , τ) = 2 πp0SN° e en 0

2 − 2 π ⋅ θ − k22θ (2) ∆a

θ dθ =

θ1

π ⋅ p0 ∆ (a2 )SN° − π ⋅ T∆rt 2 ⋅ e ⋅ e 2 a2

(

τ p −| τ | Tr

−e

τ p −| τ | − π ⋅ 2

Tr

),

(4.175) where SN is the speciﬁc effective scattering area under vertical scanning; Tr =

∆(a2 )Td ∆(a2 ) h = ; 2 c a2 4 a2

a2 = 1 +

k 2 ∆(a2 ) ; 4π

(4.176)

(4.177)

and|τ| ≤ τp . When the duration of the pulsed searching signal is short, i.e., the condition τp << Tr is satisﬁed, reference to Equation (4.175) shows that R0en (t , τ) = p(t) ⋅ (1 − |ττp|) ,

Copyright 2005 by CRC Press

(4.178)

1598_book.fm Page 195 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

195

where

p(t) =

p0 ∆ (a2 )SN π τ p − π ⋅ ∆Trt ⋅ ⋅e . Tr 2 a2

(4.179)

The normalized correlation functions of the ﬂuctuations given by Equation (4.175) and Equation (4.178) are shown in Figure 4.21 for various values of the ratio

τp Tr

. The variables t and τ are separable for these normalized corre-

lation functions, i.e., the normalized correlation function R0en (t , τ) is separable within the limits of the region given by Equation (4.170). Because the normalized correlation function R0en (t , τ) is periodic, the power spectral density is linear. The envelope of the power spectral density of the ﬂuctuations, with the accuracy of the power dependent on the time t, is deﬁned by the Fourier transform with respect to Equation (4.175) for the arbitrary pulsed signal duration, R 0en (t, τ) 1.00

0.75

1 0.50 2

3 0.25

τ τp 0

0.25

0.50

0.75

1.00

FIGURE 4.21 The normalized correlation function

R 0en

tions at n = 0 and vertical scanning: (1)

Copyright 2005 by CRC Press

τp Tr

(t, τ) of the intraperiod target return signal ﬂuctua<< 1; (2)

τp Tr

= 1.0; (3)

τp Tr

= 2.0.

1598_book.fm Page 196 Monday, October 11, 2004 1:57 PM

196

Signal and Image Processing in Navigational Systems

S (ω ) ≅ en

Under the condition

τp Tr

ch

π τp 2 Tr

− cos ω τ p

( ) +ω τ 2

π τp 2 Tr

.

(4.180)

2 2 p

→ 0, the power spectral density Sen(ω) is deﬁned by

sinc2 (0.5ωτp) and corresponds to the normalized correlation function given by Equation (4.178). The normalized power spectral density Sen(ω) given by Equation (4.180) is shown in Figure 4.22 for various values of the ratio

τp Tr

and under the condition τp << Tr . At high values of frequency, Sen(ω) is decreased in accordance with the law ω12 . Reference to Equation (4.175) shows that integrating with the variable τ, we can deﬁne the effective correlation interval as τ = 4 Tr ⋅ ef c

Under the condition

τp Tr

ch

π τp 2 Tr π τp 2 Tr

−1

π sh

.

(4.181)

→ 0, we obtain τ cef → τ p , which is what it must be.

With an increase in the value of τp , the effective correlation interval is S en (ω) 1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 3

0.2 0.1

2

1 0

1

2

3

4

ω τp 5

6

7

8

9

10

11

12

FIGURE 4.22 The power spectral density Sen(ω) of the intraperiod target return signal ﬂuctuations under vertical scanning: (1)

Copyright 2005 by CRC Press

τp Tr

<<1; (2)

τp Tr

= 1; (3)

τp Tr

= 2.

1598_book.fm Page 197 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

197

4T

increased and tends to approach the ratio π r . The zero correlation interval determined under the condition R(τ) = 0 is always equal to 2τp , as follows from Equation (4.175). Let us consider the intermediate region given by Equation (4.171) and Equation (4.172) when ∆ t ∈ [– 0.5 τp , 0.5 τp]. For this region, the correlation function is deﬁned by two functions. The ﬁrst function is the normalized correlation function of the ﬂuctuations given by Equation (4.175) that follows from Equation (4.169) under the conditions θ1 > 0 and θ2 > 0 . The second function follows from Equation (4.175) under the condition θ1 = 0 and takes the following form: R0en (t , τ) =

2 ∆ t + τp − | τ | −π⋅ p0 ∆ (a2 )SN 2Tr 1 ⋅ − e . 2 a2

[

]

(4.182)

Within the limits of the region 0 ≤ ∆ t ≤ 0.5τp given by Equation (4.171) under the condition ∆ t ≤ 0.5 |τ| ≤ 0.5τp , the correlation function of the ﬂuctuations is determined by Equation (4.175). Under the condition 0 ≤ 0.5 |τ| ≤ ∆ t, it is given by Equation (4.182). Within the limits of the interval – 0.5τp ≤ ∆ t ≤ 0 [see Equation (4.172)], Equation (4.182) is true and the variable τ is within the limits of the interval τ ∈ [0, ∆ t + 0.5τp]. Because the variables t and τ are not separable in the correlation function given by Equation (4.182), the correlation function is not separable and the instantaneous power spectral density depends essentially on the time t within the limits of the regions given by Equation (4.171) and Equation (4.172). The zero correlation intervals depend also on the time t. Under the condition t ≥ Td or ∆ t ≥ 0, the zero correlation interval is equal to τp , for the region given by Equation (4.170). If the condition t ∈ [Td – 0.5 τp , Td] is satisﬁed, we can consider that τ c = τ p + 2 ∆ t ∈[0, τ p ] ,

(4.183)

i.e., the correlation interval is less than τp , and under the condition ∆ t → – 0.5τp , the correlation interval tends to approach zero. This is clear because the instant of time ∆ t = – 0.5τp corresponds to the leading edge of the pulsed searching signal reaching the scanned surface of the two-dimensional target. At the instants of time that are within the limits of the interval [Td – 0.5τp , Td], only those parts of the pulsed searching signal can reach the scanned surface of the two-dimensional target that are removed from the leading edge of the pulsed searching signal at the time equal to ∆ t < 0.5τp. Evidently, with the shift in the time of two overlapping pulsed searching signals in the integrand in Equation (2.122) by the value ±∆ t, the product of these two pulsed searching signals becomes equal to zero and the correlation disappears. Thus, the effective instantaneous power spectral density bandwidth, which is inversely proportional to the correlation interval, tending to

Copyright 2005 by CRC Press

1598_book.fm Page 198 Monday, October 11, 2004 1:57 PM

198

Signal and Image Processing in Navigational Systems

approach ∞ at the instant of time when the pulsed searching signal reaches the scanned surface of the two-dimensional target, is rapidly decreased as the pulsed searching signal is propagated and, at the instant of time t = Td , approaches the value τ1p , which is constant. The correlation function of the ﬂuctuations is determined by Equation (4.182) within the limits of the time interval [Td , Td + 0.5τp] and is not separable, as seen earlier. The instantaneous power spectral density of the ﬂuctuations is deformed, although its effective bandwidth is approximately constant. The target return signal power increases and becomes maximal if the condition ∆ t = 0.5τp is satisﬁed (see Section 2.6). 4.8.2

The Interperiod Fluctuations with the Vertically Moving Radar

Let us consider the interperiod target return signal ﬂuctuations in the glancing radar range when the vertically moving radar approaches the scanned surface of the two-dimensional target. We assume that the conditions τ = nTp′ and ε0 = – 0.5π are true in Equation (2.182). In this case, the correlation function of the interperiod ﬂuctuations can be written in the following form: 2 π θ2

R (t , τ ) = p 0 en 0

∫ ∫ g (θ cos α − ϕ , 2

0

0

θ sin α − ψ 0 ) ⋅ S (θ) ⋅ e jΩmax cos θ dθ dα ,

θ1

(4.184) where

θ1,2 = θ*2 ∓

c τp 2h

=

2 (ρ* − h ∓ 0.25c τ p ) h

=

2( ∆ t ∓ 0.5 τ p )

∆ t = t − Td ≥ 0.5τ p .

Td

; (4.185)

(4.186)

Reference to Equation (4.185) and Equation (4.186) shows that unlike in Equation (4.173), the limits of integration θ1 and θ2 are independent of the parameter τ. For the instants of time that are within the limits of the time interval [– 0.5τp , 0.5τp], we obtain the angle θ1 = 90°. Multiplying Equation (4.184) with the factor e jω 0τ and using the Fourier transform and the ﬁltering property of the delta function, the power spectral density of the ﬂuctuations can be written in the following form: S(ω , t) ≅ f [θ(ω )] ⋅ Π[θ1 (ω , t), θ2 (ω , t)] , where Copyright 2005 by CRC Press

(4.187)

1598_book.fm Page 199 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

199

2π

f [θ(ω )] =

∫ g (θ cos α − ϕ , θ sin α − ψ ) ⋅ S°(θ) dα ; 2

0

0

(4.188)

0

cos θ =

ω −ω 0 Ωmax

, Π = 1 at θ ∈ [θ1, θ2], and Π = 0 at θ ∉ [θ1, θ2]. The boundary

values of the angles θ1 and θ2 given by Equation (4.186) deﬁne the boundary frequencies of the instantaneous power spectral density at the instant of time t = c ∗ independently of the type of the function f [θ(ω)]. Under the condition ∆ t > 0.5 τp or ρ* > h + 0.25cτp , the boundary frequencies of the power spectral density have the following form: 2ρ

ω 1,2 = ω 0 + Ωmax cos θ1,2 = ω 0 +

Ωmax h Ωmax h = ω0 + . ρ1,2 ρ* ∓ 0.25c τ p

(4.189)

If the condition 0.5c τp << h is satisﬁed, then the effective bandwidth at the boundary frequencies can be written in the following form: ∆Ω = ω 1 − ω 2 ≅

Ωmax c τ p h 2ρ

2 *

=

cτp 2h

⋅ Ωmax cos 2 θ* .

(4.190)

If the condition θ* < ∆a << 1 is true, then, in practice, under the condition ∆ t > 0.25τp , the effective power spectral density bandwidth is very slightly decreased because the average angle θ* increases and tends to approach constant. The average frequency of the power spectral density given by ω = ω 0 + Ω max cos θ *

(4.191)

is slowly decreased, also with an increase in the value of the angle θ* . To deﬁne the dependence of the parameters of the power spectral density as a function of the time, we can write cos θ = hρ*−1 = Td t −1 = Td (Td + ∆ t) −1 ≅ 1 − ∆ tTd−1 ,

(4.192)

∆ t = t − Td = 0.5c(ρ* − h) ≥ 0.5τ p .

(4.193)

where

Substituting Equation (4.192) in Equation (4.190) and Equation (4.191), we can write

Copyright 2005 by CRC Press

1598_book.fm Page 200 Monday, October 11, 2004 1:57 PM

200

Signal and Image Processing in Navigational Systems ∆Ω = 0.5c τ p h−1Ωmax (1 − ∆ tTd−1 )2

and

ω = ω 0 + Ωmax (1 − ∆ tTd−1 ) . (4.194)

The angle θ1 = 0° within the limits of the following intervals ∆ t ∈ [– 0.5τp , 0.5τp] or ρ* ∈ [h – 0.25cτp , h + 0.25cτp]. Consequently, the following equality ω1 = ω0 + Ωmax is true. In addition, we can assume ρ2 = h + 0.25cτp + 0.5c∆t, where |∆ t| = |t – Td| ≤ 0.5 τp. Then the effective bandwidth can be written in the following form: ∆Ω = Ωmax − Ωmax cos θ2 = Ωmax (1 − hρ2−1 ) = 0.25c τ p hΩmax (1 + 2 ∆t τ −p1 ) . (4.195) Under the condition ∆ t < – 0.5τp, i.e., t < Td – 0.5τp , the target return signal is as yet absent at the receiver or detector input in any navigational system. The effective bandwidth is linearly increased from zero to the value of ∆Ωmax = 0.5Ωmax c τ p h−1

(4.196)

within the limits of the time interval ∆ t ∈ [– 0.5τp , 0.5τp], and after that it is approximately constant until the target return signal disappears completely. This character of the power spectral density of the target return signal ﬂuctuations is clear. At the ﬁrst instants of time, when the pulsed searching signal could reach the scanned surface of the two-dimensional target, i.e., when an illuminated spot is very small in dimension and the difference ρ2 – h is low in value, changes in the dimensions of the scanned surface of the two-dimensional target or changes in the corresponding differences of phases of the target return signal, which are caused by the interperiod ﬂuctuations, are very low in value. This fact can explain why the power spectral density is narrow-band. At the instant of time equal to t = Td – 0.5τp , when the pulsed searching signal could reach the scanned surface of the two-dimensional target, the illuminated spot is reduced to a point. The moving radar does not produce any ﬂuctuations, i.e., the effective bandwidth is equal to zero at this instant of time. Our interest is in comparing this conclusion with the conclusion made in Section 4.8.1. The effective power spectral density bandwidth of the interperiod ﬂuctuations in the radar range was changed from inﬁnity to minimal value, and after that had held the value constant within the limits of the same time interval. It follows from the results already discussed that if the duration of the pulsed searching signal is short and the function f(ω) is smoothed, the shape of the function f(ω) deﬁnes the target return signal power as a function of time t or the angle θ* (see Section 2.6), but it does not deﬁne the shape of the power spectral density, which is close to that of a rectangle with a skewed top (see Figure 4.23). The shape of the rectangle top

Copyright 2005 by CRC Press

1598_book.fm Page 201 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

Π (t3)

Π (t2)

201

S (ω)

Π (t1)

1.0

f (ω)

S (ω, t 3)

S (ω, t 2)

S (ω, t 1) ω

FIGURE 4.23 The shape of the power spectral density of the interperiod ﬂuctuations as a function of appearance time of the target return signal under the conditions τp << Tr and θ0 ≠ 0°.

is not important for analysis if it is not spiked. The main information regarding the power spectral density is contained in the values of the effective bandwidth and average frequency. Because of this, the determination of the function f(ω) is of interest for us when the duration of the pulsed searching signal is large in value ﬁrst, and especially when the pulsed searching signal is a continuous nonmodulated process. In this case, the power spectral density of the interperiod ﬂuctuations is deﬁned by the complete function f[θ(ω)] given by Equation (4.188), as the condition given by Equation (2.109) is true. If the directional diagram is approximated by the Gaussian bottle and S°(θ) = const within the limits of the radar antenna beam width, then we can obtain the power spectral density Sg(ω) given by Equation (3.136) based on Equation (4.187) and Equation (4.188), in which ∆h and ∆v must be replaced with the notations ∆ϕ and ∆ψ , respectively, for obtaining the effective bandwidth of Sg(ω) in the angle planes ϕ and ψ. Finally, we can say that the main results discussed in Section 3.3 are true for the considered case here. Under the condition θ0 = 0°, we can obtain a simple solution if the speciﬁc 2

effective scattering area is determined in the form S°(θ) = e − k2θ . The power spectral density of the ﬂuctuations with the continuous searching signal is given by Equation (3.144), in which the values ∆ϕ and ∆ψ must be replaced with the values

∆ϕ a

and

∆ψ a

, respectively, where a = 1 +

k2 ∆(a2 ) 2π

. If the condition

∆ϕ = ∆ψ is satisﬁed, we can use the power spectral density Sg(ω) given by Equation (3.142).

Copyright 2005 by CRC Press

1598_book.fm Page 202 Monday, October 11, 2004 1:57 PM

202

Signal and Image Processing in Navigational Systems

S (ω) 1.0 s (ω, t3)

s (ω, t2)

s (ω, t1)

f (ω)

S (ω, t3)

S (ω, t2)

S (ω, t1) ω

FIGURE 4.24 The shape of the power spectral density of the interperiod ﬂuctuations as a function of appearance time of the target return signal under the conditions τp << Tr and θ0 ≠ 0°.

The instantaneous power spectral densities S(ω, t) with various time cross sections t, if the following conditions θ0 = 0°, ∆ϕ = ∆ψ, and τp << Tr are true, are shown in Figure 4.24. Thus, according to Equation (3.142), we can write f (ω ) = e

−

ω 0 + Ω max − ω ∆Ω

(4.197)

and only the power depends on the time-cross-section t, but not the shape of the power spectral density. The normalized power spectral density of the ﬂuctuations can be written in the following form: s(ω , t ) =

S(ω , t ) , S(ω , t )

(4.198)

where ω is the average frequency and is shown in Figure 4.24 by the dotted line. The normalized power spectral density s(ω, t) of the ﬂuctuations is independent of the time cross section t.

4.8.3

The Interperiod Fluctuations with the Horizontally Moving Radar

Let the radar motion be parallel to the underlying surface of the Earth, i.e., ε0 = 0. The interperiod target return signal ﬂuctuations in the glancing radar range (τ = nTp′ ) are deﬁned by the correlation function determined in the following form:50

Copyright 2005 by CRC Press

1598_book.fm Page 203 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

203

V β0

θ0 V

ψ0 ϕ0

β0

FIGURE 4.25 The position of the vector of velocity of the moving radar relative to the radar antenna directional diagram and two-dimensional (surface) target. 2 π θ2

R en (t , τ) = p0

∫ ∫ g (θ cos α − ϕ , 2

0

0

θ sin α − ψ 0 )

θ1

(4.199)

× S°(θ) ⋅ e jΩmaxτ sin θ cos(α − β0 ) sin θ dθ dα , where θ1 and θ2 are given by Equation (4.184), β0 is the angle characterizing the mutual position of the directional diagram and the vector of velocity of the moving radar (see Figure 4.25). Multiplying Equation (4.199) with the factor e jω 0τ and using the Fourier transform, the power spectral density S(ω, t) can be written in the following form: 2 π θ2

S(ω , t) ≅

∫ ∫ g (θ cos α − ϕ , θ sin α − ψ ) 2

0

0

0

θ1

(4.200)

× S°(θ)sin θ ⋅ δ[ω 0 + Ωmax sin θ cos(α − β 0 ) − ω ] dθ dα Assuming sin θ ≈ θ and introducing a new variable θ=

Copyright 2005 by CRC Press

ω − ω0 + x , Ω max cos(α − β 0 )

(4.201)

1598_book.fm Page 204 Monday, October 11, 2004 1:57 PM

204

Signal and Image Processing in Navigational Systems

we can write 2 π θ2

S(ω , t) ≅

∫ ∫ g [θ(x) cos α − ϕ , θ(x)sin α − ψ ] 2

0

0

0

θ1

× S°[θ( x)] ⋅

ω − ω0 + x 2 Ωmax cos 2 ( α − β 0 )

⋅ δ( x) dx dα =

,

α2

=

∫ g [θ(0) cos α − ϕ ,

ω − ω0 + x

θ(0)sin α − ψ 0 ] ⋅ S°[θ( x)] ⋅

2

0

Ω2max cos 2 ( α − β 0 )

α1

dα (4.202)

where the values of θ(0) =

υ cos( α −β 0 )

, υ=

ω − ω0 Ωmax

, and α 1,2 = β 0 + arccos

υ θ1 , 2

are

deﬁned under the conditions θ(0) = θ1 and θ(0) = θ2. Substituting the value of θ(0) in the integrand in Equation (4.202) and introducing a new variable y = tg (α – β0), we can write y2

S(ω , t) ≅

∫ g (υ, y) ⋅ S°(υ 2

1 + y2

) dy ,

(4.203)

y1

where g( υ, y ) = g[υ(cos β 0 − y sin β 0 ) − ϕ 0 , υ(sin β 0 + y cos β 0 ) − ψ 0 ]

(4.204)

and y1,2 = θ 12,2 υ −2 − 1. If the duration of the pulsed searching signal is short, so that the length of the interval [y1, y2] is low in value, using the theorem about the mean, we can write

(

1 S(ω , t) ≈ Ω−max υ ⋅ g 2 (υ, y* ) ⋅ S° υ 1 + y *2

) ⋅ (y

2

− y1 ) ,

(4.205)

where y∗ = θ∗2 υ −2 − 1 and y2 − y1 = υ −1  θ∗2 + 0.5 c τ p h−1 − υ 2 − 

θ∗2 − 0.5c τ p h−1 − υ 2  .  cτ

cτ

(4.206)

We must keep in mind that y1 = 0 if υ 2 ∈[θ∗2 − 2 hp , θ∗2 + 2 hp ]. If the duration of the pulsed searching signal is very large in value, i.e., the condition τp >> Tr is satisﬁed or the pulsed searching signal is a continuous process, then

Copyright 2005 by CRC Press

1598_book.fm Page 205 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

205

integration with respect to the variable y is carried out within the limits of the interval (–∞, ∞). Let us assume that the directional diagram and the speciﬁc effective scattering area S°(θ) are deﬁned by the Gaussian law. The power spectral density of the ﬂuctuations given by Equation (4.202) can be deﬁned with an arbitrary duration of the pulsed searching signal in the following form: y2

S(ω , t) ≅ Ω

−1 max

∫

υ e

 υ (cos β 0 − y sin β 0 ) − ϕ 0  − 2π ⋅ 

{

2

∆ ϕ2

 υ (sin β 0 + y cos β 0 ) − ψ 0  +  2 ∆ψ

2

} − k υ(1+ y ) 2

2

dy .

y1

(4.207) The integral in Equation (4.207) is deﬁned using the error integral given in Equation (1.27). In the general case, detailed study and analysis of this are very cumbersome.51 Because of this, we limit our consideration to particular cases. Let us suppose that the directional diagram axis is downward directed, i.e., the conditions θ0 = 0°, ϕ0 = 0°, and ψ0 = 0° are satisﬁed, and β0 = 0°. Then, we can write

[ (

S(ω , t) ≅ Φ

2 2 π aψ ( θ 22 − υ2 )

∆(ψ2 )

)− Φ (

2 2 π aψ ( θ12 − υ 2 )

∆(ψ2 )

)]⋅ e

− 2π⋅

aϕ2 ∆(2) ϕ

⋅ υ2

, (4.208)

where aϕ2 = 1 + 0.5π −1 k2 ∆ (ϕ2 )

aψ2 = 1 + 0.5π −1 k2 ∆ (ψ2 ) .

and

(4.209)

We need to compare this with Equation (4.175). With the continuous searching signal and under the condition β0 = 0°, we can write

S(ω ) ≅ e

− 2π ⋅

aϕ2 ∆(ϕ2 )

⋅

(ω − ω 0 )2 2 Ωmax

,

(4.210)

i.e., the power spectral density of the ﬂuctuations coincides with the square of the directional diagram narrowed in aϕ times on the plane of the cross section ϕ, in which there is the vector of velocity of the moving radar. The effective bandwidth of the power spectral density S(ω) can be determined in the following form: ∆F =

Copyright 2005 by CRC Press

2 V ∆ϕ ⋅ aϕ λ

(4.211)

1598_book.fm Page 206 Monday, October 11, 2004 1:57 PM

206

Signal and Image Processing in Navigational Systems

and depends on the power of roughness of the scanned surface of the twodimensional target using the coefﬁcient k2. With the scanned two-dimensional target having a very rough surface, i.e., under the condition aϕ ≈ 1, the effective bandwidth of S(ω) is maximal. As the roughness of the scanned surface of the two-dimensional target is smoothed, the effective bandwidth of S(ω) is decreased. In the short-range area, the effective bandwidth of the power spectral density S(ω) of the ﬂuctuations can be determined in the following form:

∆F = V ⋅

2 ∆(ϕ2 ) λ2 aϕ2

+

aϕ2

(4.212)

2 h 2 ∆(ϕ2 )

instead of Equation (4.211) (see Section 4.7). The ﬁrst term in Equation (4.212) deﬁnes the ﬂuctuations caused by the Doppler effect; the corresponding correlation interval is equal to the time during which the radar moves the distance equal to the diameter of the radar antenna. The second term in Equation (4.212) deﬁnes the ﬂuctuations caused by the exchange of scatterers within the limits of the illuminated spot; the corresponding correlation interval is equal to the time during which the radar moves the distance equal to the dimensions of the illuminated spot. In the long-range area, we can neglect this effect. If the duration of the pulsed searching signal is short and the condition β0 = 0° is satisﬁed, the power spectral density S(ω,t) can be written in the following form: S(ω , t) ≅

(

−1

θ + 0.5c τ p h − υ − 2 *

2

−1

θ − 0.5c τ p h − υ 2 *

2

)⋅ e

− 2π ⋅

2 2 ψ

2 2 ϕ

(2) ψ

(2) ϕ

( a∆ θ∗ + υ∆ a

−

2 υ2 aψ

∆(ψ2 )

) .

(4.213) Under the condition β0 = 90°, we have the same results, but it is necessary to replace the parameters aϕ and ∆ϕ with the parameters aψ and ∆ψ, respectively, in Equation (4.208)–Equation (4.213). When the directional diagram is axial symmetric, the parameters aϕ , ∆ϕ and aψ , ∆ψ must be replaced with the parameters aa and ∆a , respectively. Reference to Equation (4.213) shows that the power spectral density S(ω, t) of the ﬂuctuations is symmetric relative to the frequency ω0 and is within the limits of the frequency interval [ω1, ω2], where ω 1 = ω 0 − Ωmax ⋅

θ∗2 + 0.5c τ p h−1

and ω 2 = ω 0 + Ωmax ⋅

θ∗2 + 0.5c τ p h−1 . (4.214)

For this reason, the effective bandwidth of S(ω, t) at zero frequencies under the condition ∆ t ≥ – 0.5τp is determined in the following form:

Copyright 2005 by CRC Press

1598_book.fm Page 207 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

∆Ω = 2 Ωmax ⋅

θ∗2 + 0.5c τ p h−1 = 2 Ωmax ⋅ 2( ∆ t + 0.5 τ p )Td−1 .

207

(4.215)

Because the angle θ* is increased as the pulsed searching signal is propagated, the effective bandwidth of S(ω, t) depends on the radar range or time cross section t. The average frequency of the effective bandwidth of S(ω, t) is independent of the parameter t and always coincides with the parameter ω0. Reference to Equation (4.215) shows that under the condition ∆ t = – 0.5τp , the effective bandwidth ∆Ω of S(ω, t) at zero frequencies is equal to zero: ∆Ω = 0. The reason for this is choosing the origin of the coordinate t, which is to be related to the middle of the pulsed searching signal — the instant of time when the leading edge of the pulsed searching signal reaches the scanned surface of the two-dimensional target: t0 = Td – 0.5τp . The normalized power spectral density s(ω, t) given by Equation (4.213) is shown in Figure 4.26 for the axial symmetry directional diagram ∆h = ∆v with various values of the angle θ* at the relative delay 2 ⋅ ∆Trt = θ *2

Tr τp

. The

power spectral density s(ω, t) is normalized, so that under the conditions ω = ωmax and ω = ωmin, for all values θ* the power spectral densities are equal to 1: s(ω , t) ≅

∆ t τ −p1 + 0.5 − 0.5υ 2Tr τ −p1 −

∆ t τ −p1 − 0.5 − 0.5υ 2Tr τ −p1 . (4.216)

The power spectral density S(ω, t) given by the exact formula in Equation (4.208) is shown in Figure 4.26 by the dotted line. The main conclusions made in this section coincide with the conclusions presented in Jukovsky et al.52

3

2

s (ω, t) 1

1.0

0.5

Ω Ωmax −3

−2

−1

0

1

2

3

FIGURE 4.26 The power spectral density of the interperiod ﬂuctuations as a function of appearance time of the target return signal under vertical scanning of the two-dimensional (surface) target and radar moving in the horizontal way: (1) ∆τ pt = 0.5; (2) ∆τ pt = 1; (3) ∆τ pt = 2.5.

Copyright 2005 by CRC Press

1598_book.fm Page 208 Monday, October 11, 2004 1:57 PM

208

4.9

Signal and Image Processing in Navigational Systems

Determination of the Power Spectral Density

To determine the power spectral density of the target return signal Doppler ﬂuctuations we use the technique discussed in Section 3.3.3. Let us assume, for example, that the radar moves horizontally. The geometric locus with the same Doppler frequency is the line corresponding to the cross section of the Earth’s surface or the scanned surface of the two-dimensional target by the lateral area of the cone, the axis of which coincides with the vector of velocity of the moving radar (see Figure 4.27). This line is often called the isodope and is the hyperbola determined in the form h2 = x2tg2θ – γ2. The surface plane of the Earth or the two-dimensional target can be covered by the totality of these hyperbolae with the parameter θ (see Figure 4.28). The positions of these hyperbolae can be expressed by asymptotes in the form y = ± tgθ.53,54 The power of the target return signal corresponding to the interval of the Doppler frequencies [Ω, Ω + dΩ] is equal to the power of the target return signal from the narrow zone of the Earth or the scanned two-dimensional target surface placed between the hyperbolae with the parameters θ and θ + dθ (see Figure 4.29, the hatched area) within the limits of the area . If the azimuth angle β0 is sufﬁciently high by value (see Figure 4.29), the area of each narrow zone of the Earth or scanned two-dimensional target surface is the same for all hyperbolae. For this reason, the power of the target return signal from this area is deﬁned only by the coefﬁcient of ampliﬁcation of the radar antenna — both the receiving and the transmitting radar antenna — for the given direction. The power spectral density of the ﬂuctuations

V

h

y

θ C x

FIGURE 4.27 Formation of the isodope: the cross section of the cone of the constant Doppler frequencies by the Earth’s surface.

Copyright 2005 by CRC Press

1598_book.fm Page 209 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

209

y Asymptote

h

θ

x x = h ctg θ

FIGURE 4.28 Formation of the isodope: the totality of isodopes and asymptotes of isodopes.

coincides in shape with the square of the horizontal-coverage directional diagram. The same result was discussed in Section 4.5.2 [see Equation (4.118)]. If the azimuth angle β0 of the radar antenna is low in value and the aspect angle of the considered area is high in value, the hyperbolae have a large radius of curvature near the tops. Then the angle AOB (see Figure 4.30) contracting the segment of the hyperbola AB within the limits of the area is more than the beam width of the horizontal-coverage directional diagram for the majority of hyperbolae. Because of this, the power of the target return signal corresponding to the given Doppler frequency depends on the coefﬁcient of ampliﬁcation of the radar antenna in the vertical plane, and not on the shape of the horizontal-coverage directional diagram, except in the case of small areas near the boundaries of the region . The shape of the power spectral density is deﬁned, in the general case, by the shape of a part of the vertical-coverage directional diagram that covers the region . This phenomenon is deﬁned by Equation (4.68). Thus, we can rigorously determine the power spectral density of the Doppler ﬂuctuations based on the statements just discussed. Here we do not present these formulae because the obtained results coincide with the results we just deﬁned. As an exception, let us show the results for the exact determination of the power spectral density of the Doppler ﬂuctuations under the condition β0 = 0° and various values of the aspect angle, as this case was investigated in a rather simple manner. Under the condition β0 = 0° and low values of the aspect angle, the angle A′OB′ can be less than the beam width of the horizontal-coverage directional diagram (see Figure 4.30). Then the limits of integration with respect to the variable ϕ depend on the variable ψ near the boundaries of the region , and Equation (4.4), rigorously speaking, cannot Copyright 2005 by CRC Press

1598_book.fm Page 210 Monday, October 11, 2004 1:57 PM

210

Signal and Image Processing in Navigational Systems

θ + dθ

c τp

F

2 cos γ

θ

β0

β1

β2

FIGURE 4.29 Space relations for determination of the power spectral density of the target return signal Doppler ﬂuctuations with the high-deﬂected directional diagram of radar antenna.

be written in the form of the product, as in Equation (4.6). For this reason, the exact formula of the power spectral density is different from that given by Equation (4.68). With the Gaussian vertical-coverage directional diagram, the power spectral density of the Doppler ﬂuctuations takes the following form:

[ (

S(Ω) = S0 Φ

2 π Ω ∆ hΩmax

⋅

Ω2 − Ω Ω2

)− Φ (

2 π Ω ∆ hΩmax

Ω1 − Ω Ω1

⋅

)] ,

(4.217)

where S0 =

PG02 λ3 ∆ h S ( γ * ) g v2 ( γ * − γ 0 ) ; 64 π 3 h 3V

(4.218)

and Ω 1,2 = Ω max ⋅ cos γ 1,2 = Ω max ⋅

Copyright 2005 by CRC Press

h 1 − (ρ ∓ 0.25 cτ p )2 ∗ 2

(4.219)

1598_book.fm Page 211 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

211

θ +dθ θ

c τp 2 cos γ A

F

A′ O B′

β2

B

FIGURE 4.30 Space relations for determination of the power spectral density of the target return signal Doppler ﬂuctuations with the nondeﬂected directional diagram of the radar antenna.

are the Doppler frequencies corresponding to the inside and outside boundaries of the region ; Ω2 is the maximum frequency: Ω1 < Ω2. Where Ω < Ω1 , we have to use Equation (4.217) to determine the power spectral density. Where Ω1 < Ω < Ω2 , we have to assume that the second term in Equation (4.217) is equal to zero, and where Ω > Ω2 , that both terms in Equation (4.217) are equal to zero. As can be seen from Figure 4.31, the power spectral density of the Doppler ﬂuctuations given by Equation (4.217) is signiﬁcantly different from the square waveform power spectral density only at low values of the aspect angle. The power spectral density bandwidth at the level 0.5 — which in practice is the effective bandwidth — is determined by Equation (4.70). The frequency interval, within the limits of which the power spectral density is 0.1 Ω

∆( 2 )

increased from 0 to 0.9 or decreased from 1 to 0.1, is equal to cosmaxγ * h . In other words, we can say that this frequency interval decreases as the horizontal-coverage directional diagram is narrowed and the value of the aspect angle is lowered. This technique of power spectral density deﬁnition was used by Sokolov and Chadovich55 to determine the power spectral density

Copyright 2005 by CRC Press

1598_book.fm Page 212 Monday, October 11, 2004 1:57 PM

212

Signal and Image Processing in Navigational Systems

1.0

S (Ω) S0 3

0.9

2

1

0.8 0.7

a

b

c

d

0.6 0.5 0.4 0.3 0.2 3

0.1

Ω Ωmax

1

2

0

88

86

0.

0.

72

70

0.

0.

54 0. 52

50

0.

0.

36

34

0.

32

0.

30

0.

28

0.

26

0.

0.

FIGURE 4.31 The power spectral density of the target return signal Doppler ﬂuctuations with the nondecτ

ﬂected directional diagram of the radar antenna at 2 hp = 0.3 and various values of ∆h and γ1: (1) ∆h = 3°; (2) ∆h = 6°; (3) ∆h = 12°; (a) γ1 = 75°; (b) γ1 = 60°; (c) γ1 = 45°; (d) γ1 = 30°.

with the radar moving nonhorizontally and vertically scanning the twodimensional (surface) target.

4.10

Conclusions

Under scanning of the two-dimensional (surface) target by the moving radar, the total normalized correlation function Ren(t, τ) of the target return signal ﬂuctuations can be expressed as the product of two normalized correlation functions: the azimuth-normalized correlation function Rβ(t, τ) and the aspect-angle normalized correlation function Rγ(t, τ). Rβ(t, τ) takes into consideration the slow target return signal ﬂuctuations caused by the difference in the Doppler frequencies in the azimuth plane. Rγ(t, τ) takes into consideration the slow ﬂuctuations caused by the difference in the Doppler frequencies in the aspect-angle plane and the rapid ﬂuctuations, which in turn are caused by the propagation of the pulsed searching signal along the scanned surface of the two-dimensional target. While scanning the two-dimensional (surface) target by the pulsed searching signal, when the radar is stationary, the correlation function Ren[t(ψ*), τ] is a periodic function with respect to the parameter τ with the period Tp . Ren[t(ψ*), τ] deﬁnes the rapid ﬂuctuations in the radar range. Because, in the general case, the variables t(ψ*) and τ are not separable, the correlation function in the radar range is not separable, i.e., the spectral characteristics depend on the parameter τ. The corresponding instantaneous power spectral Copyright 2005 by CRC Press

1598_book.fm Page 213 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

213

density is the regulated function. The envelope of the power spectral density is deﬁned by the Fourier transform for the correlation function Ren[t(ψ*), τ] and depends on the parameter t. Under scanning of the two-dimensional (surface) target by the pulsed searching signal when the radar moves, the aspect-angle correlation function Rγ(t, τ) deﬁnes the interperiod ﬂuctuations in the glancing radar range and is not separable. The power spectral density Sγ[ω, ψ* (t)] of the interperiod ﬂuctuations in the glancing radar range depends essentially on the value of the angle ψ* . With the Gaussian pulsed searching signal, the aspect-angle normalized correlation function Rγ(τ, ψ*) is separable and is deﬁned by the product of the two normalized correlation functions Rγ′ (τ) and Rγ″ (τ, ψ*). Rγ′ (τ) deﬁnes the ﬂuctuations in the radar range. Rγ″ (τ, ψ*) deﬁnes the slow ﬂuctuations that are caused by the difference in the Doppler frequencies at the aspect-angle plane, and the interperiod ﬂuctuations in the glancing radar range. Under scanning of the two-dimensional (surface) target by the pulsed searching signal when the radar moves, the azimuth-normalized correlation function Rβ(τ) deﬁnes the target return signal ﬂuctuations caused by the difference in the Doppler frequencies of scatterers observed with various values of the azimuth angle within the horizontal-coverage directional diagram. The power spectral density Sβ(ω) corresponding to Rβ(τ) is deﬁned for two cases: the high- and low-deﬂected radar antenna. Under scanning of the two-dimensional (surface) target by the moving radar, the total normalized correlation function Ren(t, τ) of the ﬂuctuations is deﬁned by the product of the aspect-angle normalized correlation function Rγ(t, τ) and the azimuth-normalized correlation function Rβ(τ). Rγ(t, τ) deﬁnes the rapid ﬂuctuations in the radar range and the slow interperiod ﬂuctuations caused by the difference in the radial components of the velocity of scatterers in the aspect-angle plane. Rβ(τ) deﬁnes only the interperiod ﬂuctuations caused by difference in the radial components of the velocity of scatterers in the azimuth plane. For this reason, the product of Rγ(t, τ) and Rβ(τ) leads to changes only in the interperiod ﬂuctuations and does not act on ﬂuctuations in the radar range. The interperiod ﬂuctuations in the glancing radar range are deﬁned by the product of the aspect-angle normalized correlation function Rγ(t, τ) and the azimuth-normalized correlation function Rβ(τ). The total power spectral density in the glancing radar range is deﬁned by the convolution between the aspect-angle power spectral density Sγ[ω, ψ*(t)] and azimuth power spectral density Sβ(ω). The total normalized correlation function in the ﬁxed radar range is deﬁned by the product of Rγ(t, τ) under the condition τ = nTp and Rβ(τ). In the general case, we have to use various numerical techniques to deﬁne the power spectral densities Sβ(ω) and Sγ(ω). In considering of the short-range area, the normalized correlation function and power spectral density are deﬁned by the Gaussian law, as in the case of the long-range area. The difference is only in the effective bandwidths, between ∆F1 and ∆F by the additional term decreasing with an increase in Copyright 2005 by CRC Press

1598_book.fm Page 214 Monday, October 11, 2004 1:57 PM

214

Signal and Image Processing in Navigational Systems

the radar range ρ. This additional term takes into consideration changes in the amplitudes of elementary signals when the scatterers are covered by the directional diagram and is deﬁned by the time of scanning the scatterers by the radar antenna beam, i.e., the time when the scatterers are within the limits of the radar antenna beam width. The value of the effective bandwidth ∆F1, unlike the value of the effective bandwidth ∆F, is inversely proportional to the directional diagram width or the radar antenna beam width. Under vertical scanning of the two-dimensional (surface) target, three cases are possible: the radar is stationary; the radar moves only vertically; the radar moves only horizontally. If the radar is stationary, there are only the intraperiod ﬂuctuations. In this case, the correlation function R0en (t , τ) is periodic with respect to the variable τ. The power spectral density is the regulated function. When the radar moves only vertically, there are the interperiod ﬂuctuations in the glancing radar range. The effective power spectral density bandwidth is linearly increased from zero to ∆Ωmax within the limits of the time interval [– 0.5τp , 0.5τp], and after that it is approximately constant until the target return signal disappears completely. If the radar moves only horizontally, the interperiod ﬂuctuations in the glancing range are deﬁned by the correlation function Ren(t, τ) given by Equation (4.199) and the power spectral density S(ω, t) given by Equation (4.200).

References 1. Trump, T. and Ottersten, B., Estimation of nominal direction of arrival and angular spread using an array of sensors, Signal Process., Vol. 50, No. 2, 1996, pp. 57–70. 2. Li, J., Liu, G., Jiang, N., and Stoica, P., Moving target feature extraction for airborne high-range resolution phased-array radar, IEEE Trans., Vol. SP-49, No. 2, 2001, pp. 277–289. 3. Katkovnik, V., A new concept of beamforming for moving sources and impulse noise environment, Signal Process., Vol. 80, No. 4, 2000, pp. 1863–1882. 4. Farina, A., Antenna-Based Signal Processing Techniques for Radar Systems, Artech House, Norwood, MA, 1992. 5. Ward, J., Space-time adaptive processing for airborne radar, Technical Report, 1015, Lincoln Laboratory, MIT, Cambridge, December 1994. 6. Abramovich, Yu, Spencer, N., and Gorokhov, A., Detection-estimation of more uncorrelated gaussian sources than sensors in non-uniform linear antenna arrays — part 1: Fully augmentable arrays, IEEE Trans., Vol. SP-49, No. 5, 2001, pp. 959–971. 7. Gerlach, K. and Steiner, M., Adaptive detection of range distributed targets, IEEE Trans., Vol. SP-47, No. 7, 1999, pp. 1844–1851. 8. Jacobs, S. and O’Sullivan, A., High-resolution radar models for joint tracking and recognition, in Proceedings of the 1997 IEEE National Radar Conference, Syracuse, NY, May 1997, pp. 99–104. 9. Scharf, L., Statistical Signal Processing, Addison-Wesley, Reading, MA, 1991. Copyright 2005 by CRC Press

1598_book.fm Page 215 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

215

10. Johnson, D. and Dudgeon, D., Array Signal Processing: Concepts and Techniques, Prentice Hall, Englewood Cliffs, NJ, 1993. 11. Moffet, A., Minimum-redundancy linear arrays, IEEE Trans., Vol. AP-16, No. 2, 1968, pp. 172–175. 12. Besson, O., Vincent, F., Stoica, P., and Gershman, A., Approximate maximum likelihood estimators for array processing in multiplicative noise environments, IEEE Trans., Vol. SP-48, No. 9, 2000, pp. 2506–2518. 13. Dedersen, K., Mogensen, P., and Flenry, B., Analysis of time, azimuth and Doppler dispersion in outdoor radio channels, in Proceedings of the ACTS Mobile Communications Summit, Aalbory, Denmark, Oct. 1997, pp. 308–313. 14. Besson, O. and Stoica, P., Decoupled estimation of DOA and angular spread for a spatially distributed source, IEEE Trans., Vol. SP-48, No. 7, 2000, pp. 1872–1882. 15. Davies, R., Brennan, H., and Reed, I., Angle estimation with adaptive arrays in external noise ﬁelds, IEEE Trans., Vol. AES-12, No. 3, 1976, pp. 179–186. 16. Reid, D., Zoubir, A., and Boashash, B., Aircraft ﬂight parameter estimation based on passive acoustic techniques using the polynomial Wigner–Ville distribution, J. Acoust. Soc. Amer., Vol. 102, No. 1, 1997, pp. 207–223. 17. Ward, K., Baker, F., and Watts, S., Maritime surveillance radar — Part 1: Radar scattering from the ocean surface, in Radar and Signal Processing, IEE Proceedings F, Vol. 137, No. 2, April 1990, pp. 330–340. 18. Conte, E., Di Bisceglie, M., Galdi, C., and Ricci, G., A procedure for measuring the coherence length of the sea texture, IEEE Trans., Vol. IM-46, No. 4, 1997, pp. 836–841. 19. Gerlach, K. and Steiner, M., Fast converging adaptive detection of Dopplershifted range distributed targets, IEEE Trans., Vol. SP-48, No. 9, 2000, pp. 2686–2690. 20. Kullback, S., Information Theory and Statistics, Dover, Mineola, NJ, 1997. 21. Gardner, W., Statistical Spectral Analysis: A Non-Probability Theory, Prentice Hall, Englewood Cliffs, NJ, 1988. 22. Mendel, J., Lessons in Estimation Theory for Signal Processing, Communications, and Control, Prentice Hall, Englewood Cliffs, NJ, 1995. 23. Raich, R., Goldberg, J., and Messer, H., Bearing estimation for a distributed source via the conventional beamformer, in Proceedings of the SSAP Workshop, Portland, OR, September 1988, pp. 5–8. 24. Telatar, E., Capacity of multi-antenna Gaussian channels, Europ. Trans. Telecommun., Vol. 10, No. 6, 1997, pp. 585–596. 25. Mandurovsky, I., Spectral performances of target return signals from the Earth surface under presence of side-lobes, Problems in Radio Electronics, Vol. OT, No. 9, 1978, pp. 3–12 (in Russian). 26. Dickey, F., Labitt, M., and Staudaher, F., Development of airborne moving target radar for long-range surveillance, IEEE Trans., Vol. AES-27, No. 11, 1991, pp. 959–971. 27. Unghes, P., A high-resolution range radar detection strategy, IEEE Trans., Vol. AES-19, No. 9, 1983, pp. 663–667. 28. Schleher, D., MTI and Pulsed Doppler Radar, Artech House, Norwood, MA, 1991. 29. Farina, A., Scannapieco, F., and Vinelli, F., Target detection and classiﬁcation with very high range resolution radar, in Proceedings of the International Conference on Radar, Versailles, France, April 1989, pp. 20–25.

Copyright 2005 by CRC Press

1598_book.fm Page 216 Monday, October 11, 2004 1:57 PM

216

Signal and Image Processing in Navigational Systems

30. Jeruchim, M., Balaban, P., and Shanmugam, K., Simulation of Communication Systems, Plenum, New York, 1992. 31. Stuber, G., Principles of Mobile Communication, Kluwer, Norwell, MA, 1996. 32. Pahlavan, K. and Levesque, A., Wireless Information Networks, John Wiley & Sons, New York, 1995. 33. Robertson, P. and Kaisen, S., The effects of Doppler spreads in OFDM (A) mobile radio system, in Proceedings of the Vehicular Technology Conference (VTC’99-Fall), 1999, pp. 329–333. 34. Einarsson, G., Principles of Lightwave Communications, John Wiley & Sons, New York, 1996. 35. Stoica, P. and Moses, R., Introduction to Spectral Analysis, Prentice Hall, Upper Saddle River, NJ, 1997. 36. Grifﬁths, L. and Jim, C., An alternative approach to linearly constrained adaptive beamforming, IEEE Trans., Vol. AP-30, No. 1, 1982, pp. 27–34. 37. Tzvetkov, A., Non-Stationary Stochastic Processes, Energy, Moscow, 1973 (in Russian). 38. Rappaport, T., Wireless Communications Principle and Practice, Prentice Hall, Englewood Cliffs, NJ, 1996. 39. O’Sullivan, J., De Vore, M., Kedia, V., and Miller, I., Automatic target recognition performance for SAR imagery using a conditionally Gaussian model, IEEE Trans., Vol. AES-37, No. 1, 2001, pp. 91–108. 40. Hill, D. and Bodie, J., Carrier detection of PSK signals, IEEE Trans., Vol. COM49, No. 3, 2001, pp. 487–495. 41. Lance, E. and Kaleh, G., A diversity scheme for a phase-coherent frequencyhopping spread-spectrum system, IEEE Trans., Vol. COM-45, No. 9, 1997, pp. 1123–1129. 42. Aschwartz, M., Bennet, W., and Stein, S., Communication System and Techniques, IEEE Press, New York, 1996. 43. Proakis, J., Digital Communications, 3rd ed., McGraw-Hill, New York, 1995. 44. Feldman, Yu, Fluctuations of the target return signal caused by moving radar, Problems in Radio Electronics, Vol. OT, No. 6, 1972, pp. 3–21 (in Russian). 45. Mensa, D., High-Resolution Radar Cross Section Imaging, Artech House, Norwood, MA, 1991. 46. Wicker, S., Error Control Systems for Digital Communication and Storage, Prentice Hall, Englewood Cliffs, NJ, 1995. 47. Stoica, P., Besson, O., and Gershman, A., Direction-of-arrival estimation of an amplitude-distorted wave-front, IEEE Trans., Vol. SP-49, No. 2, 2001, pp. 269–276. 48. Kolchinsky, V., Mandurovsky, I., and Konstantinovsky, M., Doppler Apparatus and Navigational Systems, Soviet Radio, Moscow, 1975 (in Russian). 49. Simon, M., Hinedi, S., and Lindsey, W., Digital Communication Technique, Prentice Hall, Englewood Cliffs, NJ, 1995. 50. Wehner, D., High Resolution Radar, Artech House, Norwood, MA, 1987. 51. Bowman, A. and Azzalini, A., Applied Smoothing Techniques for Data Analysis, Oxford University Press, Oxford, U.K., 1997. 52. Jukovsky, A., Onoprienko, E., and Chijov, V., Theoretical Foundations of Radar Altimetry, Soviet Radio, Moscow, 1979 (in Russian). 53. Pillai, S., Bar-Ness, Y., and Haber, F., A new approach to array geometry for improved spatial spectrum estimation, in Proceedings of the IEEE, Vol. 73, Oct 1985, pp. 1522–1524.

Copyright 2005 by CRC Press

1598_book.fm Page 217 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

217

54. Feldman, Yu, Determination of power spectral density of target return signals, Problems in Radio Electronics, Vol. OT, No. 6, 1959, pp. 22–38 (in Russian). 55. Sokolov, M. and Chadovich, I., Characteristics of Doppler spectrum under inclined ﬂight of airborne, News of the USSR University, Radio Electronics, No. 12, 1975, pp. 61–66 (in Russian).

Copyright 2005 by CRC Press

1598_book.fm Page 219 Monday, October 11, 2004 1:57 PM

5 Fluctuations Caused by Radar Antenna Scanning

5.1

General Statements

If the radar is stationary, radar antenna scanning is the main source of slow target return signal ﬂuctuations.1,2 Let us investigate the correlation function and power spectral density of the target return signal ﬂuctuations caused by the moving radar antenna in scanning the three-dimensional (space) and two-dimensional (surface) targets. Two of the most universally adopted forms of radar antenna scanning3 are line scanning used, as a rule, in the detection of targets and conical scanning used in the tracking of moving targets. Conical radar antenna scanning with simultaneous rotation of radar antenna polarization plane will also be studied here.

5.1.1

The Correlation Function under Space Scanning

If under space scanning by the radar antenna, the position of the polarization plane is changed simultaneously with the moving radar antenna, it can be assumed that ∆ξ ≠ 0. The value of ∆ξ is the same for all scatterers. Using Equation (2.75)–Equation (2.77), Equation (2.84), Equation (2.85), and Equation (2.93)–Equation (2.95), and assuming that ∆ = ∆ω = ∆ζ = 0, we can write the space–time normalized correlation function of the target return signal ﬂuctuations in the following form:4 R∆enϕ , ∆ψ , ∆ξ (t , t + τ) = Rp (t , t + τ) ⋅ Rsc ( ∆β 0 , ∆γ 0 ) ⋅ Rq ( ∆ξ) ,

(5.1)

where ∞

Rp (t , t + τ) = N ⋅

∑ ∫ P [t −

2ρ c

] ⋅ P* [t − 2cρ + τ − nTp ]ρ−2 dρ

(5.2)

n= 0

219 Copyright 2005 by CRC Press

1598_book.fm Page 220 Monday, October 11, 2004 1:57 PM

220

Signal and Image Processing in Navigational Systems

is the normalized correlation function of the ﬂuctuations in the radar range, which is caused by propagation of the pulsed searching signal; Rsc ( ∆β 0 , ∆γ 0 ) = N ⋅

∫∫ g(ϕ, ψ) ⋅ g(ϕ + ∆β

0

cos γ 0 , ψ + ∆γ 0 ) dϕ dψ

(5.3)

is the normalized correlation function of the space ﬂuctuations, which is caused by radar antenna scanning; Rq ( ∆ξ) = N ⋅

∫∫ q(ξ, ζ) ⋅ q(ξ + ∆ξ, ζ)sin dξ dζ

(5.4)

is the normalized correlation function of the space ﬂuctuations, which is caused by the rotation of the radar antenna polarization plane. With the pulsed searching signal, we can neglect variations in the variable ρ in the denominator of the integrand of the normalized correlation function Rp(t, t + τ) of the time ﬂuctuations in the radar range, which is determined by Equation (5.2), and a new variable can be introduced: z = t − 2ρ c −1 .

(5.5)

In this case, Rp(t, t + τ), which is given by Equation (5.2), differs from that given by Equation (3.7) only by the absence of compression in the time scale, i.e., µ = 1, which happens with the moving radar. In terms of what was just discussed, all the main statements and conclusions made in Section 3.1 regarding the ﬂuctuations in the radar range are true for the case considered here. With a simple searching harmonic signal, the normalized correlation function Rp(t, t + τ) becomes equal to 1. A study of the normalized correlation functions Rsc(∆β0, ∆γ0) and Rq(∆ξ) of the slow (space) ﬂuctuations, which are determined by Equation (5.3) and Equation (5.4) and caused by radar antenna scanning and rotation of the polarization plane, respectively, is of prime interest. If the variables ϕ and ψ can be separated in the function g(ϕ, ψ) — the radar antenna directional diagram — we can write Rsc ( ∆β 0 , ∆γ 0 ) = Rsch ( ∆β 0 ) ⋅ Rscv (∆γ 0 ) ,

(5.6)

where Rsch ( ∆β 0 ) = N ⋅

Copyright 2005 by CRC Press

∫ g (ϕ) ⋅ g (ϕ + ∆β cos γ ) dϕ ; h

h

0

0

(5.7)

1598_book.fm Page 221 Monday, October 11, 2004 1:57 PM

Fluctuations Caused by Radar Antenna Scanning

Rscv ( ∆γ 0 ) = N ⋅

221

∫ g (ψ) ⋅ g (ψ + ∆γ ) dψ . v

v

0

(5.8)

If we are able to write the directional diagram g(ϕ, ψ), as the function g(ϕ ′ , ψ ′) for a new coordinate system ϕ′ and ψ′, which is rotated with respect to the previous coordinate system ϕ and ψ by the angle α = arctg

∆γ 0 ∆β 0 cos γ 0

,

(5.9)

then, instead of Equation (5.3) we can write Rsc(∆β0, ∆γ0) in the following form: Rsc ( ∆β 0 , ∆γ 0 ) = N ⋅

∫∫ g(ϕ′, ψ ′) ⋅ g(ϕ′ + ∆ϕ′, ψ ′) dϕ′dψ ′,

(5.10)

where ∆ϕ ′ = ∆β 20 cos 2 γ 0 + ∆γ 20 .

(5.11)

If the variables ϕ′ and ψ′ are separable in the function g(ϕ ′ , ψ ′), i.e., g(ϕ ′ , ψ ′) = g1 (ϕ ′) ⋅ g2 (ψ ′) ,

(5.12)

Rsc(∆β0, ∆γ0) can be written in the following form: Rsc ( ∆β 0 , ∆γ 0 ) = N ⋅

∫ g (ϕ′) ⋅ g (ϕ′ + ∆ϕ′) dϕ′ , 1

1

(5.13)

where g1 (ϕ ′) is the directional diagram by power in the plane in which the directional diagram axis moves. Equation (5.3) and Equation (5.6) allow us to determine the normalized correlation function Rsc(∆β0, ∆γ0) with an arbitrary orientation of the directional diagram. For instance, in the case of the Gaussian directional diagram, Rsc(∆β0, ∆γ0) takes the following form:

Rsc ( ∆β 0 , ∆γ 0 ) = e

− 0.5π

2 0

2

[ ∆β ∆cos γ (2) h

0

+

∆γ 20 ∆(v2 )

]

.

(5.14)

In the case of the directional diagram with the uniform distribution law, for instance, the sinc2-directional diagram [see Equation (2.103)], Rsc(∆β0, ∆γ0) has the following form:

Copyright 2005 by CRC Press

1598_book.fm Page 222 Monday, October 11, 2004 1:57 PM

222

Signal and Image Processing in Navigational Systems

Rsc ( ∆β 0 , ∆γ 0 ) =

36(1 − sinc X )(1 − sinc Y ) , X 2Y 2

(5.15)

where X = 2π ⋅

∆β 0 cos γ 0 ∆h

Y = 2π ⋅

and

∆γ 0 . ∆v

(5.16)

If radar antenna scanning is given as a function of time, we can transform Rsc(∆β0, ∆γ0) into the normalized correlation function of the time ﬂuctuations without any problems. For this purpose, we need to deﬁne the angle shifts ∆β0 and ∆γ0 as a function of τ for the stationary stochastic target return signal, or as a function of t and τ for the nonstationary stochastic target return signal. With the parametric form of deﬁnition of scanning law β0 = fh(t) and γ0 = fv(t), we can write ∆β 0 = fh (t + 0.5 τ) − fh (t − 0.5 τ)

and

∆γ 0 = fv (t + 0.5 τ) − fv (t − 0.5 τ) . (5.17)

In the case of the uniform radar antenna scanning, we can write β 0 = Ωhsc ⋅ t ; γ 0 = Ωvsc ⋅ t ; ∆β 0 = Ωhsc ⋅ τ ;

∆γ 0 = Ωvsc ⋅ τ , (5.18)

and

where Ωhsc and Ωvsc are the angular velocities in the horizontal and vertical planes of radar antenna scanning, respectively. In this case, the slow ﬂuctuations can be represented as a stationary stochastic process.

5.1.2

The Correlation Function under Surface Scanning

Consider the correlation function R∆enρ, ∆ϕ , ∆ψ (t , τ) of the target return signal ﬂuctuations, which is determined by Equation (2.122). Let us assume that ∆ρ = 0 ;

∆ϕ = ∆ϕ sc = ∆β 0 cos(ψ + γ 0 ) ;

and

∆ψ = ∆ψ sc = ∆γ 0 . (5.19)

R∆enρ, ∆ϕ , ∆ψ (t , τ) can be written in the following symmetric form: R∆ϕ , ∆ψ (t , τ) = p(t) ⋅ R∆enϕ , ∆ψ (t , τ) ⋅ e jω 0τ ,

Copyright 2005 by CRC Press

(5.20)

1598_book.fm Page 223 Monday, October 11, 2004 1:57 PM

Fluctuations Caused by Radar Antenna Scanning

223

where the normalized correlation function R∆enϕ , ∆ψ (t , τ) is determined by ∞

R∆enϕ , ∆ψ (t , τ) = N ⋅

∑ ∫∫ P t –

2 ρ( ψ ) c

n= 0

× P* t – 

2 ρ( ψ ) c

− 0.5 (τ − nTp )  

+ 0.5 (τ − nTp )  

× g[ϕ − 0.5 ∆β 0 cos(ψ + γ 0 ), ψ − 0.5 ∆γ 0 ]

(5.21)

× g[ϕ + 0.5 ∆β 0 cos(ψ + γ 0 ), ψ + 0.5 ∆γ 0 ] × S°(ψ + γ 0 )sin(ψ + γ 0 ) dϕ dψ where the power p(t) of the target return signal is given by Equation (2.136). In the general case, R∆ϕ,∆ψ(t, τ) given by Equation (5.20) cannot be expressed as the product of the correlation function of the ﬂuctuations in the radar range and that of the ﬂuctuations caused by radar antenna scanning, unlike the normalized correlation function R∆enϕ , ∆ψ , ∆ξ (t , t + τ) given by Equation (5.1). This phenomenon can be explained by the fact that the radar range ρ becomes a function of the angle ψ, not an independent coordinate. However, with the short pulsed searching signals (the pulse duration is low in value), we can assume that S°(ψ + γ 0 ) ≈ S°( γ * )

sin(ψ + γ 0 ) ≈ sin γ * .

and

(5.22)

Let us introduce a new variable given by Equation (5.5) that leads us to exchange the variable ψ for the variable ψ* + c*z. Here, we neglect variations in the directional diagram g(ϕ, ψ) as a function of the parameter z within the limits of the pulsed searching signal period. Then, the normalized correlation function R∆enϕ , ∆ψ (t , τ) of the ﬂuctuations can be written in the following form: R∆enϕ , ∆ψ (t , τ) = Rp (τ) ⋅ Rsc ( ∆β 0 , ∆γ 0 ) ,

(5.23)

where ∞

Rp (τ) = N ⋅

∑ ∫ P [z − 0.5(τ − nT )] ⋅ P [z + 0.5(τ − nT )] dz ; ∗

p

n= 0

Copyright 2005 by CRC Press

p

(5.24)

1598_book.fm Page 224 Monday, October 11, 2004 1:57 PM

224

Signal and Image Processing in Navigational Systems ϕ2

Rsc ( ∆β 0 , ∆γ 0 ) = N ⋅

∫ g ϕ − 0.5∆β

0

cos γ * , − 0.5∆γ 0 

ϕ1

.

(5.25)

× g ϕ + 0.5 ∆β 0 cos γ * , ψ * + 0.5 ∆γ 0  dϕ Thus, in scanning the two-dimensional (surface) target by the pulsed searching signal, R∆enϕ , ∆ψ (t , τ) can be expressed as the product of the periodic normalized correlation function Rp(τ) of the ﬂuctuations in the radar range (the comb function), which is given by Equation (5.24), and Rsc(∆β0, ∆γ0) of the slow ﬂuctuations, which is given by Equation (5.25). These normalized correlation functions are caused by radar antenna scanning. The normalized correlation function Rp(τ) given by Equation (5.24) is an analog of Rp(t, t + τ) given by Equation (5.2). Rsc(∆β0, ∆γ0) given by Equation (5.25) depends on the angle ψ* as a function of the parameter if the shape of the directional diagram in the plane of the angle ψ is different for various values of ψ* . If the variables in the function g(ϕ, ψ) are separable, then the normalized correlation function of the slow ﬂuctuations Rsc(∆β0, ∆γ0) caused by radar antenna scanning can be expressed as the product given by Equation (5.6), where ϕ2

Rsch ( ∆β 0 ) = N ⋅

∫ g [ϕ − 0.5∆β h

0

cos γ * ] ⋅ gh [ϕ + 0.5 ∆β 0 cos γ * ] dϕ ;

(5.26)

ϕ1

Rscv ( ∆γ 0 ) =

gv (ψ * − 0.5 ∆γ 0 ) ⋅ gv (ψ * + 0.5 ∆γ 0 ) . gv2 (ψ * )

(5.27)

The normalized correlation function Rsch ( ∆β 0 ) given by Equation (5.26) is different from that determined by Equation (5.7) in the symmetric form of writing and the use of the angle γ* instead of the angle γ0. Comparison of the normalized correlation function Rscv ( ∆γ 0 ) given by Equation (5.27) with the one determined by Equation (5.8) shows the essential distinction: an integration with respect to the variable ψ is not carried out in Rscv ( ∆γ 0 ) given by Equation (5.27) because the interval of variation of the variable ψ is low in value. When the directional diagram is Gaussian, Rsc(∆β0, ∆γ0) of the slow ﬂuctuations is determined by Equation (5.14) based on Equation (5.26) and Equation (5.27), in which the angle γ* is used instead of the angle γ0. When the continuous searching signal, i.e., the condition given by Equation (2.109), is true, Rsc(∆β0, ∆γ0) follows from Equation (5.21) and can be written in the following form:

Copyright 2005 by CRC Press

1598_book.fm Page 225 Monday, October 11, 2004 1:57 PM

Fluctuations Caused by Radar Antenna Scanning

225

ϕ2 ψ 2

Rsc ( ∆β 0 , ∆γ 0 ) = N ⋅

∫ ∫ g ϕ − 0.5∆β

0

cos(ψ + γ 0 ), ψ − 0.5 ∆γ 0 

ϕ1 ψ 1

× g ϕ + 0.5 ∆β 0 cos(ψ + γ 0 ), ψ + 0.5 ∆γ 0 

.

(5.28)

× S°(ψ + γ 0 )sin(ψ + γ 0 ) dϕ dψ 5.1.3

The General Power Spectral Density Formula

Consider the normalized correlation function Rsc(∆β0, ∆γ0) of the slow target return signal ﬂuctuations, which is determined by Equation (5.13). Let the radar antenna scanning be uniform in the plane of the angle ϕ′ with the angular velocity Ωsc . Then, we can write:5 ϕ ′ = Ωsc ⋅ t

∆ϕ ′ = Ωsc ⋅ τ .

and

(5.29)

In this case, the power spectral density S(ω) of the target return signal ﬂuctuations can be written in the following form: 2

∞

∫ g (Ω t) ⋅ e 1

p(t) S(ω ) = ⋅ π

sc

− jωt

dt

−∞

.

∞

(5.30)

∫ g (Ω t) dt 2 1

sc

−∞

The formula in Equation (5.30) has a simple physical meaning. This is the square of the power spectral density of the amplitude of the target return signal, which is the Fourier transform of the individual elementary signal covering all elementary scatterers by the directional diagram during radar antenna scanning. This formula corresponds to a general physical process of forming the power spectral density of the pulsed stochastic process (the target return signal) discussed in Section 2.1. The directional diagram by power g(Ωsct), included in the formula in Equation (5.30), is the squared directional diagram by voltage or is the product of the directional diagrams by voltage for the transmitting and receiving conditions. For this reason, the amplitude power spectral density of the ﬂuctuations caused by radar antenna scanning can be considered as the Fourier transform of the squared directional diagram by voltage which, in its turn, can be expressed as the convolution between functions that are the Fourier transform of the ﬁrst order directional diagram by voltage. Let us denote this function by E(y). The function E(y) deﬁnes a distribution of the electromagnetic ﬁeld over a radar antenna area. Because of this, the

Copyright 2005 by CRC Press

1598_book.fm Page 226 Monday, October 11, 2004 1:57 PM

226

Signal and Image Processing in Navigational Systems

amplitude power spectral density of the ﬂuctuations caused by radar antenna scanning can be determined in the following form: ∞

s(ω ) =

∫ E(y) ⋅ E(ω − y) dy ,

(5.31)

−∞

where y = 2 πλ −1Ωsc x

(5.32)

and x is the distance from the center of the radar antenna area. The power spectral density of the ﬂuctuations is equal to the squared amplitude power spectral density s(ω) given by Equation (5.31).

5.2

Line Scanning

5.2.1

One-Line Circular Scanning

Let us assume that the radar antenna rotates uniformly around its vertical axis with the angular velocity Ωhsc and the constant aspect angle γ0. In doing so, ∆γ 0 = 0

∆β 0 = Ωhsc ⋅ τ .

and

(5.33)

We assume that the variables ϕ and ψ in the function g(ϕ, ψ) are separable. Then, the normalized correlation function Rsch ( ∆β 0 ) of ﬂuctuations caused by radar antenna scanning is deﬁned by scanning the two-dimensional (surface) target in the same manner as in scanning the three-dimensional (space) target [see Equation (5.7) and Equation (5.26)]. In the case of the Gaussian directional diagram, it follows from Equation (5.14) that Rsc (τ) = e

2 − πτ

τ c2

,

(5.34)

where

τc =

Copyright 2005 by CRC Press

2 ∆h Ω cos γ 0 h sc

(5.35)

1598_book.fm Page 227 Monday, October 11, 2004 1:57 PM

Fluctuations Caused by Radar Antenna Scanning

227

is the correlation interval; Ωhsc =

2π = 2 π Nsc ; Tsc

(5.36)

Tsc is the period of radar antenna scanning; and Nsc = T1 is the rotation per sc second. Using Equation (5.34), we can deﬁne the power spectral density in the following form:

Sscen (ω ) ≅ e

−π⋅

ω2 ( 2 π ∆Fsc )2

,

(5.37)

1 Ωh cos γ 0 = sc τ sc 2 ∆h

(5.38)

where ∆Fsc =

is the effective bandwidth of the power spectral density. For example, at ∆h = 2°, γ0 = 0°, and Ωsc = 1 radian , we obtain that ∆Fsc = 20 Hz. Reference to sec Equation (5.35) and Equation (5.38) shows that when the horizontal-coverage directional diagram is wide, and its angular velocity is low in value, the power spectral density is narrow, i.e., the effective bandwidth is low in value. In the case of the sinc-directional diagram, it follows from Equation (5.15) that6 Rsch (τ) = 6 −2 (1 − sinc ) ,

(5.39)

where

=

3πτ τc

τc =

and

3∆ h . 2Ω cos γ 0 h sc

(5.40)

With the low values of τ, the normalized correlation function Rsch (τ) given by Equation (5.39) is very close to the Gaussian function [see Equation (5.34) and Figure 5.1]. Using the Fourier transform for Rsch (τ) given by Equation (5.39), the power spectral density Sscen (ω ) can be determined in the following form:  [1 −  S (ω ) ≅   0 en sc

Copyright 2005 by CRC Press

|ω | 2 3 π ∆Fsc

]

at

| ω | ≤ 3 π ∆Fsc ;

at

| ω | > 3 π ∆Fsc ,

(5.41)

1598_book.fm Page 228 Monday, October 11, 2004 1:57 PM

228

Signal and Image Processing in Navigational Systems

Rsc (τ) 1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 τ τc

0.1 0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

FIGURE 5.1 Normalized correlation functions with line circular scanning for the sinc- (solid line) and Gaussian (dotted line) directional diagrams of the radar antenna.

where ∆Fsc =

1 2 Ωhsc cos γ 0 = τc 3 ∆h

(5.42)

is the effective bandwidth of Sscen (ω ) . Sscen (ω ) given by Equation (5.41) is shown in Figure 5.2 by the solid line. It has clear bounds; the extreme frequencies are equal to ω ext = ± 3 π ∆Fsc = ±

2 π Ωhsc cos γ 0 . ∆h

(5.43)

The formula in Equation (5.43) is to ﬁt a physical representation with respect to the limitation of the power spectral density Sscen (ω ) of ﬂuctuations caused by radar antenna scanning. Let us consider that with radar antenna scanning, Sscen (ω ) is formed due to the Doppler shift in the frequency of the elementary emitters of the radar antenna, which move with different angular velocities.7 The frequency ωext corresponds to emitters that are the most distant from the center of radar antenna area. Any directional diagram of a practicable radar antenna, which is a result of the Fourier transform of the electromagnetic ﬁeld distribution law given within the limits of the radar antenna area, will have Sscen (ω ) limited by frequencies ω ext = 4 π Vmax λ −1 ,

Copyright 2005 by CRC Press

(5.44)

1598_book.fm Page 229 Monday, October 11, 2004 1:57 PM

Fluctuations Caused by Radar Antenna Scanning

1.0

229

Ssc (ω)

0.8 0.7 0.6 0.5 0.4 0.3 0.2 ω − ω0 2π∆F

0.1 −1.6 −1.4 −1.2 −1.0 −0.8 −0.6 −0.4 −0.2 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4

FIGURE 5.2 Power spectral densities with line circular scanning for the sinc- (solid line) and Gaussian (dotted line) directional diagrams of the radar antenna.

where Vmax , the maximum velocity of moving radar antenna, ends with respect to the scanned target. This consequence follows immediately from Equation (5.31) and Equation (5.43). With the Gaussian directional diagram, the power spectral density Sscen (ω ) is unlimited. The Gaussian directional diagram is impracticable as it would be subject to the Gaussian distribution law of the electromagnetic ﬁeld within the limits of the radar antenna area, which is impracticable due to the ﬁnite linear dimensions of the radar antenna. This effect appears more clearly if the directional diagram is approximated by a rectangle in the following form: ω∆

Sscen (ω ) ≅ sinc 2 ( 2 Ωsc cosh γ 0 ) .

(5.45)

Thus, consideration of side lobes is very important for the deﬁnition of the power spectral density of ﬂuctuations Sscen (ω ) of the target return signal, both with moving radar (see Section 3.2.4) and under radar antenna scanning.8,9 However, if in the ﬁrst case we do not take into consideration the side lobes, the power spectral density Sscen (ω ) deprives the remainders existing in practice. If we do the same in the second case, the false remainders of the power spectral density Sscen (ω ) can appear. Under radar antenna scanning, the presence of side lobes in the directional diagram used in practice prompts us to smooth the stochastic process (the target return signal). Comparing Equation (5.35) and Equation (5.40), one can see that with the same effective width of the directional diagram, the correlation intervals and the bandwidths of the power spectral densities, respectively, differ from each

Copyright 2005 by CRC Press

1598_book.fm Page 230 Monday, October 11, 2004 1:57 PM

230

Signal and Image Processing in Navigational Systems

other by 6% — as 1.5 and 2 — despite the power spectral densities being very different in the peripheral region. The formulae in Equation (5.37) and Equation (5.41) are the power spectral densities of ﬂuctuations caused by radar antenna scanning in the form of continuous power spectral densities. However, we do not take into account the periodic iteration of the same elementary signals with each revolution of the radar antenna. Because of this, rigorously speaking, we must use the periodic sequence of elementary signals as w(t), not the elementary signal per one revolution of the radar antenna gh (Ωhsc t) : ∞

w(t) =

∑ g [Ω h

h sc

(t − nTsc )] ,

(5.46)

n= 0

where Tsc is the period of radar antenna scanning. In uniform radar antenna scanning, the normalized correlation function of the slow ﬂuctuations can be given in the following form (see Figure 5.3a): ∞

Rsch (τ , nTsc ) =

∑ R (τ − nT ) , h sc

(5.47)

sc

n= 0

where Rsch (τ) is given by Equation (5.7) if ∆β 0 = Ωhsc τ , and by Equation (5.34) and Equation (5.39), following from Equation (5.7). As a result, the power spectral density of the slow ﬂuctuations caused by radar antenna scanning is a regulated function, not continuous, with a distance between lines equal to ∆Fsc = T1 = Nsc in hertz (see Figure 5.3b) sc ∞

Ssc (ω ) = Sscen (ω ) ⋅

∑

∞

δ(ω − nΩhsc ) =

n= 0

∑ S (nΩ h sc

h sc

) ⋅ δ(ω − nΩhsc ) ,

(5.48)

n= 0

where Sscen (ω ) is given by Equation (5.37) or Equation (5.41). For instance, Ssc (ω ) ≈ e

−π⋅

∞

ω2 (2 π ∆Fsc )2

⋅

∑ δ(ω − nΩ

h sc

).

(5.49)

n= 0

With the continuous searching signal, the process of one-line scanning generates a stationary periodic stochastic process (the target return signal). The normalized correlation function Rsch (τ , nTsc ) of the slow ﬂuctuations caused by radar scanning, which is given by Equation (5.47), and the power spectral density Sscen (ω ) given by Equation (5.48) result in an exhaustive representation regarding spectral properties of the target return signal.

Copyright 2005 by CRC Press

1598_book.fm Page 231 Monday, October 11, 2004 1:57 PM

Fluctuations Caused by Radar Antenna Scanning

231

Rsc (τ)

τ 0

Tsc

2Tsc

(a) Ssc (ω)

ω 0 Ωsc (b) FIGURE 5.3 The normalized correlation function (a) and power spectral density (b) with the continuous searching signal — line scanning by the radar antenna.

With the pulsed searching signal, as follows from Equation (5.1), R (τ , nTsc ), which is determined by Equation (5.47), must be multiplied with Rp(t, t + τ) of ﬂuctuations in the radar range, which is given by Equation (5.2). The normalized correlation function Rp(t, t + τ) and the corresponding power spectral density are shown in Figure 5.4. The normalized correlation function of ﬂuctuations caused by radar antenna line scanning, which is obtained as the product of Equation (5.2) and Equation (5.47), and the corresponding power spectral density deﬁned by the convolution of the power spectral densities given by Equation (3.15) and Equation (5.48), are shown h sc

in Figure 5.5 under the condition m =

Tsc Tp

, where m is integer. Thus, the sto-

chastic target return signal is rigorously periodic, but not stationary, because the normalized correlation function of the rapid ﬂuctuations can be varied within the limits of the pulsed searching signal duration Tp , as discussed in Section 4.3, and the period of the normalized correlation function is equal to Tsc = τ. Copyright 2005 by CRC Press

1598_book.fm Page 232 Monday, October 11, 2004 1:57 PM

232

Signal and Image Processing in Navigational Systems

Rp (τ)

Tp >> Tsc

τ Tp

0

2Tp (a) Sp (ω) Ωsc << Ωp

ω − Ωp

Ωp

0 (b)

FIGURE 5.4 The normalized correlation function (a) and power spectral density (b) with the continuous searching signal — no line scanning by the radar antenna.

If the ratio

m1 m2

=

Tsc Tp

is not an integer and is a rational number, then the

normalized correlation function of rapid ﬂuctuations is not rigorously periodic, and its values will be modulated with the period equal to m2 = Tsc. Thus, the power spectral density will have discrete components with frequencies multiple to

Ωsc m2

. The degree of modulation and values of these com-

ponents are high, and the number of pulsed signals per beam is less. If the ratio

Tsc Tp

is a rational number, there are continuous remainders near the main

harmonics of the power spectral density of the ﬂuctuations. As a rule, the process of radar antenna scanning is slow, so that the distance between spectral lines of the power spectral densities given by Equation (5.37) and Equation (5.41) is proportional from 10–800 Hz. Because of this, in many cases, we can neglect the regulated character of the power spectral density and assume that one is continuous, making simpler the

Copyright 2005 by CRC Press

1598_book.fm Page 233 Monday, October 11, 2004 1:57 PM

Fluctuations Caused by Radar Antenna Scanning

233

Rp (τ) . Rsc (τ)

τ 0 Tp

Tsc

2Tsc

(a) Ssc (ω) * Sp (ω)

ω − Ωp

Ωp

0 (b)

FIGURE 5.5 The normalized correlation function (a) and power spectral density (b) with the pulsed searching signal — line scanning by the radar antenna.

determination and computer calculation without high error. A different situation arises when the power spectral density is estimated with azimuthtracking radar antenna scanning. These antennas function at about 100 r/ sec.10 In this case, we cannot neglect the regulated character of the power spectral density of the ﬂuctuations (see Section 5.3).

5.2.2

Multiple-Line Circular Scanning

Let us assume that the radar antenna performs more complex (multiple-line or “spiral”) scanning of space. For the case considered here, it can be written that β 0 = Ωhsc ⋅ t ; γ 0 = Ωvsc ⋅ t ; ∆β 0 = Ωhsc ⋅ τ ; and ∆γ 0 = Ωvsc ⋅ τ .

(5.50)

If the variables ϕ and ψ are separable in the function g(ϕ, ψ), Equation (5.6)–Equation (5.8) are true. Substituting Equation (5.50) in Equation Copyright 2005 by CRC Press

1598_book.fm Page 234 Monday, October 11, 2004 1:57 PM

234

Signal and Image Processing in Navigational Systems

(5.6)–Equation (5.8) and taking into consideration the character of radar antenna scanning with respect to the angle β0 [see Equation (5.47)], the normalized correlation function Rsc(τ) of the slow target return signal ﬂuctuations can be written in the following form: ∞

Rsc (τ) = Rscv (τ) ⋅

∑ R (τ − nT ) . h sc

h sc

(5.51)

n= 0

The formula in Equation (5.51) is true for any relationship between Ωhsc and Ωvsc . First, we assume that Ωhsc >> Ωvsc and the horizontal-coverage and vertical-coverage directional diagrams have a width with the same order. Then, as follows from Equation (5.35) and Equation (5.40), we can assume that τ vc >> τ hc , where τ hc is the correlation interval of the ﬂuctuations with the moving radar in the horizontal plane; τ vc is the correlation interval of the ﬂuctuations with the moving radar in the vertical plane. The normalized correlation function Rsch (τ) is periodic (see Figure 5.7a) and the normalized correlation function Rscv (τ) is not periodic (see Figure 5.6a) with respect to the variable τ. Because of the directional diagram shift with respect to the aspect angle, an “interrevolution” correlation is broken from revolution to revolution of radar antenna scanning, and the target return signal becomes the nonperiodic process (see Figure 5.8a). If the value of Ωvsc is increased, the value of τ vc is decreased. Under the condition τ vc < τ hc , the “interrevolution” correlation is completely broken during one revolution of radar antenna scanning.11 Therefore, the sum in Equation (5.51) has only a single term under the condition n = 0. The power spectral density of the ﬂuctuations contains only a single domain near the zero frequency. With the continuous searching signal, the previously discussed normalized correlation functions of the ﬂuctuations and the corresponding power spectral densities (see Figure 5.6b–Figure 5.8b) give us full information about the properties of the frequency of the target return signal. With the pulsed searching signal, it is necessary to multiply the normalized correlation function Rsc(τ) given by Equation (5.51) with the normalized correlation function Rp(t, t + τ) of ﬂuctuations in the radar range given by Equation (5.2) (see Figure 5.9a). In this case, the resulting power spectral density of the ﬂuctuations is the convolution between the power spectral densities Ssc(ω) and Sp(ω) shown in Figure 5.9b. This is similar to the power spectral density shown in Figure 5.7b, but instead of discrete δ components with the frequency

1 Tsch

,this power spectral density is formed by narrow partial domains

with the bandwidth equal approximately to

Copyright 2005 by CRC Press

1 τ vc

(see Figure 5.10).

1598_book.fm Page 235 Monday, October 11, 2004 1:57 PM

Fluctuations Caused by Radar Antenna Scanning

235

v

R s c (τ)

τ 0 (a) Svs c (ω)

ω 0 (b) FIGURE 5.6 The normalized correlation function (a) and power spectral density (b) with the continuous searching signal — multiple-line (“spiral”) scanning by the radar antenna in the aspect-angle plane.

5.2.3

Line Segment Scanning

Let us consider the case when the radar antenna, the directional diagram axis of which is under the constant angle γ0 , performs the line scanning in a segment limited by the angles ±βm . Let us assume that the radar antenna rotates according to the law β 0 = β m sin Ω at ,

(5.52)

where Ω a = 2Taπ and Ta is the period of radar antenna hunting. Using Equation (5.17), we can ﬁnd that ∆β 0 = 2β m sin 0.5Ω a τ cos Ω a t

Copyright 2005 by CRC Press

and

∆γ 0 = 0

(5.53)

1598_book.fm Page 236 Monday, October 11, 2004 1:57 PM

236

Signal and Image Processing in Navigational Systems

R hsc (τ)

τ 0

T hsc

2 T hsc (a) S hsc (ω)

ω − Ω hsc

Ω hsc

0 (b)

FIGURE 5.7 The normalized correlation function (a) and power spectral density (b) with the continuous searching signal — multiple-line (“spiral”) scanning by the radar antenna in the azimuth plane.

or, using Equation (5.52) and excluding the time in an explicit form, we can write ∆β 0 = 2 β m2 − β 20 (t) sin 0.5Ω a τ .

(5.54)

Reference to Equation (5.54) shows that the observed target return signal is a nonstationary periodic stochastic process because the normalized correlation function of the target return signal ﬂuctuations used in Equation (5.53) will be a periodic function of time unlike the normalized correlation function with uniform circular radar antenna scanning, when the target return signal is also periodic but stationary because the normalized correlation function is independent of time. This is clearly under consideration, for instance, for the Gaussian directional diagram and substitution of Equation (5.53) and Equation (5.54) in Equation (5.14). Then, we can write the normalized correlation function Rsc(t, τ) of the ﬂuctuations in the following form: Copyright 2005 by CRC Press

1598_book.fm Page 237 Monday, October 11, 2004 1:57 PM

Fluctuations Caused by Radar Antenna Scanning

237

Rsc (τ) = Rvsc (τ) . R hsc (τ)

τ 0

T hsc

2T hsc (a) v

h

Ssc (ω) = S sc (ω) * S sc (ω)

ω −

Ω hsc

0

Ω hsc

(b) FIGURE 5.8 The normalized correlation function (a) and power spectral density (b) with the continuous searching signal — multiple-line (“spiral”) scanning by the radar antenna in the aspect-angle and azimuth planes.

Rsc (t ,τ) = e

− 2π

β 2m ∆(h2 )

cos 2 γ 0 sin 2 0.5 Ω aτ cos 2 Ω at

=e

−2 π

β 2m − β 20 ( t ) ∆(h2 )

cos 2 γ 0 sin 2 0.5 Ω aτ

.

(5.55)

Rsc(t, τ) of ﬂuctuations caused by radar antenna scanning, which is given by Equation (5.55), is a periodic function with respect to the parameter τ with ﬁxed values of t or β0(t). In this case, t or β0(t) is the parameter, and the function of t is also periodic. Rsc(t, τ), which is determined by Equation (5.55), is shown in Figure 5.11 with various values of the parameter L that can be given by the following form: L=

Copyright 2005 by CRC Press

β 2m − β 20 (t) cos 2 γ 0 . ∆ (h2 )

(5.56)

1598_book.fm Page 238 Monday, October 11, 2004 1:57 PM

238

Signal and Image Processing in Navigational Systems

R p (τ)

τ 0

Tp

2Tp (a) Sp (ω)

ω − Ωp

Ωp

0 (b)

FIGURE 5.9 The normalized correlation function (a) and power spectral density (b) with the pulsed searching signal — no multiple-line (“spiral”) scanning by the radar antenna.

Ssc (ω) * Sp (ω)

ω − Ωp

− Ωhsc 0

Ωhsc

Ωp

FIGURE 5.10 The power spectral density with the pulsed searching signal — multiple-line (“spiral”) scanning.

The formula in Equation (5.55) is true for any relationships between the width of the scanned segment 2βm and the horizontal-coverage directional diagram width, even if the condition βm < ∆h is satisﬁed. In particular, if the

Copyright 2005 by CRC Press

1598_book.fm Page 239 Monday, October 11, 2004 1:57 PM

Fluctuations Caused by Radar Antenna Scanning

239

L1 > L2 > L3 > L4 = 0

Rsc (τ)

L4

L3 L1

L2

τ 0

Ta

2Ta

FIGURE 5.11 The correlation function — line segment scanning by the radar antenna.

radar antenna is stationary, i.e., the condition βm = 0 is satisﬁed, the clear results follow from Equation (5.55): Rsc(t, τ) 1. However, in the subsequent discussion, we assume that the condition βm >> ∆h is satisﬁed, as it is carried out in practice. Then, integrating Equation (5.55) with respect to the parameter τ within the limits of the scanning period Ta of the radar antenna, i.e., when the condition Ta >> τc is satisﬁed, we can determine the correlation interval of the observed target return signal in the following form: τ c = Ta ⋅ e − π L(t ) I0 [ π L(t )] ,

(5.57)

which is a function of the parameter β0(t). The bandwidth ∆Fsc = τ1c of the instantaneous power spectral density of the ﬂuctuations caused by radar antenna scanning is also a function of the moving position of radar antenna, i.e., β0(t). Reference to Equation (5.57) shows that under the conditions β0 << βm , and βm >> ∆h , the correlation interval can be determined in the following form: τc =

Ta π 2 L(t)

=

1 2 . ⋅ Ωa L(t)

(5.58)

Deﬁning the parameter β0 as a function of t and using Equation (5.52), we can easily show that the correlation interval τc is equal to the time during which the directional diagram axis is moved in the angle equal to the effective width of the squared horizontal-coverage directional diagram. The physical meaning of this result is obvious. Using the Fourier-series expansion for the periodic normalized correlation function Rsc(t, τ) of the ﬂuctuations, which is determined by Equation (5.55), with respect to the modiﬁed Bessel function given by Equation (1.15), we can obtain the instantaneous power spectral density Ssc(ω) in the following form: Copyright 2005 by CRC Press

1598_book.fm Page 240 Monday, October 11, 2004 1:57 PM

240

Signal and Image Processing in Navigational Systems ∞

Ssc (ω ) ≈ 2 π ⋅ e − π⋅L(t )

∑ I [π L(t)] δ(ω − kΩ ) . k

a

(5.59)

k=0

Ssc(ω) is a regulated function. Amplitudes of harmonics depend on the angle β0(t), i.e., Ssc(ω) is deformed continuously during radar antenna scanning. In particular, if the radar antenna is stationary, i.e., the condition βm = 0 is satisﬁed, or under the condition β0 = βm ≠ 0, it follows from Equation (5.59) that the power spectral density Ssc(ω) of the ﬂuctuations can be determined in the form Ssc(ω) ≈ δ(ω) that corresponds to the condition Rsc(τ) 1 obtained from the conditions in Equation (5.55). Because the normalized correlation function Rsc(τ) is different from zero if the conditions |τ| ≤ τc and τc << Ta are satisﬁed, we can write sin 0.5Ω a τ ≅ 0.5Ω a τ

(5.60)

and, using Equation (5.54), we obtain ∆β 0 ≅ β mΩ a τ cos Ω at = Ω a β 2m − β 20 (t) τ ,

(5.61)

where βmΩa cos Ωat is the instantaneous velocity of radar antenna scanning. Exchanging the function ∆β0(t) given by Equation (5.53) within the limits of the interval that is of interest to us by the segment of the tangent given by Equation (5.61) implies that we can neglect the variations in the radar antenna angular velocity within the limits of the correlation interval if the condition ∆a << βm is satisﬁed. In this case, the target return signal is the quasistationary process. Substituting Equation (5.61) in Equation (5.14) or Equation (5.15) under the condition ∆γ0 = 0, we obtain the formulae coinciding with Equation (5.34) or Equation (5.39), in which τc =

2 ka Ω a L(t)

,

(5.62)

where ka = 1 in Equation (5.34) or ka = 23 in Equation (5.39). Naturally, Equa2 tion (5.62) takes the same form as Equation (5.58). Hereafter, we can use all results discussed in Section 5.2.1 for the case of one-line radar antenna scanning and, in particular, we can deﬁne the instantaneous power spectral densities of the ﬂuctuations, which coincide with the formulae in Equation (5.37), Equation (5.41), Equation (5.48), and Equation (5.49), but the effective bandwidth of which is equal to ∆Fsc = τ c−1 , where the correlation interval τc given by Equation (5.62) is a function of the angle β0(t). For example, in the case of the Gaussian directional diagram, it follows from Copyright 2005 by CRC Press

1598_book.fm Page 241 Monday, October 11, 2004 1:57 PM

Fluctuations Caused by Radar Antenna Scanning

241

Equation (5.48) that the power spectral density Ssc(ω) of the ﬂuctuations can be written in the following form:

Ssc (ω ) ≈ e

−

∞

ω2 2 π Ω2a L ( t )

⋅

∑

∞

δ(ω − kΩ a ) =

∑

δ(ω − kΩ a ) ⋅ e

−

k2 2 π L( t )

.

(5.63)

k=0

k=0

We can prove without any problem that under the condition L(t) >> 1, the power spectral densities of the ﬂuctuations given by Equation (5.59) and Equation (5.63) are close in shape. Using the discussed technique in this section, we can investigate other laws of radar antenna scanning that are different from the law given by Equation (5.52).

5.2.4

Line Circular Scanning with Various Directional Diagrams under Transmitting and Receiving Conditions

Consider the case when the directional diagrams are different under transmitting and receiving conditions. Then, the function g(ϕ, ψ) can be replaced with the product of the directional diagrams given by Equation (2.73). First, consider a situation when one of the directional diagrams, for instance, under the transmitting condition, is nondirected, i.e., the condition gut (ϕ , ψ ) ≡ 1 is satisﬁed. In this case, the normalized correlation function Rsch ( ∆β 0 ) of the target return signal ﬂuctuations can be determined in the following form: Rsch ( ∆β 0 ) = N ⋅

∫g

r uh

r (ϕ) ⋅ guh (ϕ + ∆β 0 cos γ 0 ) dϕ

(5.64)

instead of Equation (5.7). Assume that the distribution law of electromagnetic ﬁeld within the limits of the radar antenna area is uniform, so that r guh (ϕ) = sinc ( π∆ rϕ ) ,

(5.65)

where ∆ r is the horizontal-coverage directional diagram width under the receiving condition. In the case of the sinc-directional diagram, the effective horizontal-coverage directional diagram width is the same both in power and in voltage despite the width of the main lobe being different. Taking into consideration the condition ∆β0 = Ωscτ, the normalized correlation function Rsch (τ) can be determined in the following form: Rsch (τ) = sinc (

Copyright 2005 by CRC Press

π Ωsc cos γ 0τ ∆r

).

(5.66)

1598_book.fm Page 242 Monday, October 11, 2004 1:57 PM

242

Signal and Image Processing in Navigational Systems

The power spectral density of ﬂuctuations corresponding to Rsch (τ) , given by Equation (5.66), is the square waveform power spectral density within the limits of the interval [− π Ωsc∆cos γ 0 , r

π Ωsc cos γ 0 ∆r

] . The power spectral density band-

width of the ﬂuctuations is determined in the following form: ∆F =

Ωsc ∆r

⋅ cos γ 0 .

(5.67)

It is easy to deﬁne how the form of the directional diagram acts on the shape of the power spectral density Ssc(ω) under the transmitting condition. If the distribution law of electromagnetic ﬁeld within the limits of the transmitting radar antenna area is uniform, Ssc(ω) has the shape of a trapezium (see Figure 5.12). The width of the bottom base is given by Equation (5.67) and the width of the top base is determined in the following form: Ω

∆F ′ = [ ∆sc − r

Ωsc ∆t

] ⋅ cos γ 0 ,

(5.68)

where ∆t is the width of the directional diagram by power under the transmitting condition. If the condition ∆t >> ∆r is satisﬁed, we can neglect the stimulus of the transmitting radar antenna area and consider the previous case. If the condition ∆t << ∆r is true and the directional diagrams are the same under the transmitting and receiving conditions according to Equation (5.63), we can assume that ∆F′ = 0 and the trapezoidal power spectral density Ssc(ω) is Ssc (ω)

ω − π∆F π∆F ′

0

π∆F ′ π∆F

FIGURE 5.12 The power spectral density with line scanning by the radar antenna for the different transmitting and receiving directional diagrams of the radar antenna.

Copyright 2005 by CRC Press

1598_book.fm Page 243 Monday, October 11, 2004 1:57 PM

Fluctuations Caused by Radar Antenna Scanning

243

converted into the triangular one [see Equation (5.41)]. If under transmitting and receiving conditions the directional diagrams are deﬁned by the Gaussian law and have the different effective widths by power ∆ t and ∆r , respectively, then the normalized correlation function Rsc(τ) and the power spectral density Ssc(ω) are determined by Equation (5.34) and Equation (5.37), respectively, in which the parameter ∆r must be replaced with the parameter 2 ∆ r ∆t ∆(r2 ) + ∆(t 2 )

5.3

.

Conical Scanning

Consider a classical one-beam one-channel navigational system with the conical radar antenna scanning of open and hidden forms from a large number of various navigational systems with the target tracking by direction.13,14 For navigational systems with conical radar antenna scanning of the open type, the directional diagrams under transmitting and receiving conditions are the same and matched, and move in space continuously. In this case, an essential disadvantage is the transmitting radar antenna scanning because information regarding frequency and phase of conical radar antenna scanning is known and can be used to generate deliberate interference. The radar with a hidden frequency of antenna scanning is used with the purpose of hiding the law under which radar antenna scanning is only for the receiving directional diagram. If the dimensions of the transmitting and receiving radar antennas are comparable, the radar with a hidden frequency of antenna scanning has a lesser slope of bearing characteristic and requires formation of two independent directional diagrams — stationary for the transmitting radar antenna and scanning for the receiving radar antenna. As a rule, navigational systems with target tracking use the pulsed searching signals that solve the problem of deﬁnition of distance to the target. Under the stationary condition, there is range tracking by navigational systems, and only the interperiod ﬂuctuations exist at the point of input of the receiver or detector.

5.3.1

Three-Dimensional (Space) Target Tracking

Let the radar antenna scanning be the open conical scanning with the angle θsc and frequency Ωsc (see Figure 5.13). Then we can write γ 0 = γ eq + θsc sin α 0

and

β 0 = β 0 cos γ 0 = θsc cos α 0 ,

(5.69)

where γeq is the aspect angle of equisignal direction. Under this law of radar antenna scanning, based on Equation (5.17), we can write

Copyright 2005 by CRC Press

1598_book.fm Page 244 Monday, October 11, 2004 1:57 PM

244

Signal and Image Processing in Navigational Systems

~

β0

A

α0

γ0

θsc

β0

γp

FIGURE 5.13 The coordinate system with conical scanning by the radar antenna.

∆γ 0 = 2θsc sin 0.5 ∆α 0 cos α 0

and ∆β 0 = ∆β 0 cos γ 0 = 2θsc sin 0.5 ∆α 0 sin α 0 , (5.70)

where α 0 = Ωsc ⋅ t

∆α 0 = Ωsc ⋅ τ .

and

(5.71)

Substituting Equation (5.70) and Equation (5.71) in Equation (5.14) and assuming that the directional diagram is deﬁned by the two-dimensional Gaussian function, we can write the normalized correlation function Rsc[τ, α0(t)] of the target return signal ﬂuctuations in the following form:

Rsc [τ , α 0 (t)] = e

sin 2 Ωsc t cos 2 Ωsc t −2 π θ2sc ⋅ + sin 2 0.5 Ωsc τ

(

∆(h2 )

∆(v2 )

)

.

(5.72)

Under the condition ∆h ≠ ∆v , Rsc[τ, α0(t)] given by Equation (5.72) depends on the parameter α0(t). The target return signal is the nonstationary stochastic process because the directional diagram has no axial symmetry and takes different positions with various values of α0 with respect to the direction of the moving radar producing a periodic change in the power spectral density bandwidth of the ﬂuctuations. The formula in Equation (5.72) is similar to that in Equation (5.55) and allows us to deﬁne the instantaneous power Copyright 2005 by CRC Press

1598_book.fm Page 245 Monday, October 11, 2004 1:57 PM

Fluctuations Caused by Radar Antenna Scanning

245

spectral density of the ﬂuctuations. However, unlike in Section 5.2.3, we are here interested in the power spectral density that is averaged during the period of radar antenna scanning because the target return signal is ﬁltered, and the frequency of radar antenna scanning is deﬁned. After averaging with respect to the parameter α0(t), we can deﬁne the averaged normalized correlation function Rsc [τ , α 0 (t)] in the following form: 2

Rsc [τ , α 0 (t)] = I 0 [π θ2sc b sin 2 0.5Ωsc τ] ⋅ e − π θsc a sin

2

0.5 Ωsc τ

,

(5.73)

where a = ∆ −h ( 2 ) + ∆ −v ( 2 )

b = ∆ −h ( 2 ) − ∆ −v ( 2 ) .

and

(5.74)

Rigorously speaking, it is necessary to average the correlation function of the ﬂuctuations, not the normalized correlation function. However, in this case, it is of no importance because the power of the target return signal under the scanning of the uniform three-dimensional (space) target is independent of the parameter α0 . If the directional diagram has axial symmetry, i.e., the condition ∆h = ∆v = ∆a is satisﬁed, Rsc(t, τ), based on Equation (5.72), can be written in the following form:

Rsc (t , τ) = Rsc (τ) = e

−

2 π θ2sc ∆(a2 )

sin 2 0.5 Ωsc τ

.

(5.75)

In this case, the target return signal is the stationary stochastic process. The normalized correlation functions Rsc (t , τ) and Rsc(t,τ) given by Equation (5.73) and Equation (5.75), respectively, are periodic functions with respect to the variable τ that indicates the periodic character of the averaged target return signal. For this reason, the power spectral density of the ﬂuctuations of the averaged target return signal is the regulated function with distance between harmonics equal to Ωsc. However, the Fourier transform of the averaged normalized correlation function Rsc (t , τ) of the ﬂuctuations, which is given by Equation (5.73), is not determined exactly. We can determine it with some accuracy if we employ a series expansion for the Bessel function I0(x) used in Equation (5.73), assuming that the argument is very low in value. Because the value of the parameter π θ2sc b is low, even in the case of the high axial asymmetric directional diagram, it is sufﬁcient to be limited by one or two terms of the series expansion. For instance,

Copyright 2005 by CRC Press

1598_book.fm Page 246 Monday, October 11, 2004 1:57 PM

246

Signal and Image Processing in Navigational Systems ∞

∑ N ⋅ δ(ω − kΩ

Ssc (ω ) ≈

k

sc

),

(5.76)

k=0

where N k = I k ( x) − ( 4y )2 [ I k −1 ( x) + I k +1 ( x)] + ( 8y )2 [ I k − 2 ( x) + I k +2 ( x)] ; x = 0.5π θ2sc a

and

y = 0.5π θ2sc b .

) ]

and

(5.77) (5.78)

As a rule, θsc ∈ ∆a

[(

4π

) ,( −1

−1

2π

∆h ≅ ∆v .

(5.79)

If the directional diagram is axial symmetric, i.e., the conditions a = 2 ∆ −v ( 2 )

and

b=0

(5.80)

are satisﬁed, then the power spectral density Ssc(ω) of the ﬂuctuations can be written in the following form: ∞

Ssc (ω ) ≈

∑I [ k

k=0

π θ2sc ∆(a2 )

] ⋅ δ(ω − kΩsc ) .

(5.81)

Under the condition θ∆sc = 0.3 that is characteristic for a conical scanning of a the radar antenna, the power of the ﬁrst harmonic of the power spectral density of the ﬂuctuations is for about 13% from the zero harmonic, and the power of the second harmonic is approximately 1% from the zero harmonic. For this case, the power spectral density of the ﬂuctuations caused by radar antenna scanning consists approximately of three components — a main component and two side components. Therefore, we can neglect the power of the other components of the power spectral density of ﬂuctuations of the target return signal. Axial asymmetry of the directional diagram leads to some increase in the power of harmonics, but this increase becomes significant only if the axial asymmetry of the directional diagram is high. Now consider the navigational system with the hidden scanning of the radar antenna. Let us assume that the directional diagram of the transmitting radar antenna is stationary and the directional diagram of the receiving radar antenna performs a conical scanning with the frequency Ωsc. For this case, the two-dimensional directional diagram g(ϕ, ψ) can be expressed as the product of the directional diagram gut (ϕ , ψ ) of the transmitting radar antenna

Copyright 2005 by CRC Press

1598_book.fm Page 247 Monday, October 11, 2004 1:57 PM

Fluctuations Caused by Radar Antenna Scanning

247

Equisignal Direction

Receiving Directional Diagram

Transmitting Directional Diagram Receiving Directional Diagram

FIGURE 5.14 Radar antenna directional diagrams with hidden conical scanning.

and the directional diagram gur (ϕ , ψ ) of the receiving radar antenna shifted by the angles β 0 and γ0 [see Equation (5.69) and Figure 5.14] g(ϕ , ψ ) = gut (ϕ , ψ ) ⋅ gur (ϕ + β 0 , ψ + γ 0 ) ,

(5.82)

where the angles ϕ and ψ are reckoned from the directional diagram axis of the transmitting radar antenna. Let us substitute Equation (5.82) in Equation (5.3), and take into consideration the fact that the transmitting radar antenna directional diagram is stationary. Then, the normalized correlation function Rsc(∆β0, ∆γ0) of the ﬂuctuations takes the following form: Rsc ( ∆β 0 , ∆γ 0 ) = N ⋅

∫∫ [g (ϕ, ψ)] t u

2

⋅ gur (ϕ + β 0 , ψ + γ 0 )

.

(5.83)

× g (ϕ + β 0 + ∆β 0 cos γ 0 , ψ + γ 0 + ∆γ 0 ) dϕ dψ r u

The formula in Equation (5.83) is true for any arbitrary shape of the transmitting and receiving radar antenna directional diagrams. If the one-dimensional directional diagram is deﬁned by the axial symmetric Gaussian functions, then the ﬁnite result can be obtained immediately by replacing the parameter θsc in Equation (5.72), Equation (5.73), Equation (5.75), Equation (5.76), and Equation (5.81) with the parameter 0.5θsc.12

Copyright 2005 by CRC Press

1598_book.fm Page 248 Monday, October 11, 2004 1:57 PM

248 5.3.2

Signal and Image Processing in Navigational Systems Two-Dimensional (Surface) Target Tracking

If the radar antenna performing conical scanning searches the two-dimensional (surface) target (see Figure 5.15), and the condition ∆p << ∆v is satisﬁed, it is necessary to use Equation (5.20), Equation (5.21), and Equation (5.23)–Equation (5.27), Equation (5.69), and Equation (5.70), replacing the parameter γ0 with the parameter γ* for the purpose of deﬁning the correlation function of the target return signal ﬂuctuations caused by radar antenna scanning. With the Gaussian directional diagram and under open radar antenna scanning, the correlation function Rscen (t , τ) of the ﬂuctuations caused by radar antenna scanning has the following form: Rscen (t , τ) = p(t) ⋅ Rscen (t , τ) ,

(5.84)

where the normalized correlation function Rscen (t , τ) is determined by Equation (5.72), and the power p(t) of the target return signal is given by Equation (2.138), in which

gv ( γ 0 − γ ∗ ) = e

−π

( γ p + θsc cos Ωsc t − γ ∗ )2 ∆(v2 )

.

(5.85)

Target return signal ﬂuctuations are a nonstationary and nonseparable stochastic process. This is true because both the normalized correlation function Rsc[τ, α0(t)] of the ﬂuctuations caused by radar antenna scanning [which is determined by Equation (5.72)] and the power p(t) of the target return signal depend on the parameter t, due to the amplitude modulation of the O

θsc

Gp

G

h γp

γ0 C

.A γ*

Ωsct

c τp

c τp

2cos γp 2cos γ0 FIGURE 5.15 Two-dimensional (surface) target tracking: OC — the equisignal direction; OA — the axis of the radar antenna directional diagram; Gp and G — the gated range element.

Copyright 2005 by CRC Press

1598_book.fm Page 249 Monday, October 11, 2004 1:57 PM

Fluctuations Caused by Radar Antenna Scanning

249

received target return signal gated in the radar range, which is scattered from the two-dimensional (surface) target.13,14 In the case of the axial symmetric directional diagram, when Rsc[τ, α0(t)] caused by radar antenna scanning is independent of the parameter t [see Equation (5.75)], the target return signal is a nonstationary but separable stochastic process. In this case, the averaged power spectral density of the ﬂuctuations caused by radar antenna scanning coincides with the instantaneous one if we do not consider the power of the target return signal and it has the same shape as in the case of scanning the three-dimensional (space) target [see Equation (5.81)].

5.4

Conical Scanning with Simultaneous Rotation of Polarization Plane

If under scanning the scatterers oriented chaotically in space, the position of the polarization plane is changed simultaneously with the radar antenna moving, we can assume that ∆ξ = const and ∆ζ = 0. Let us deﬁne the normalized correlation function Rq(∆ξ) of the target return signal space ﬂuctuations caused by rotation of the polarization plane of the radar antenna, which is determined by Equation (5.4), for the case when half-wave dipoles are scatterers. In this case, with consistent polarization at the transmitting and receiving conditions, we can write15 q(ξ , ζ) = cos 2 ξ sin 2 ζ .

(5.86)

Then, the normalized correlation function Rq(∆ξ, ∆ζ) of the space ﬂuctuations caused by rotation of the polarization plane simultaneously with the moving radar antenna can be written in the following form: Rq ( ∆ξ , ∆ζ) = Rq ( ∆ξ) ≈ 0.33(2 + cos 2 ∆ξ) .

(5.87)

If the polarization plane is rotated simultaneously with the angular velocity Ωsc , then ∆ξ = Ωpτ, and the normalized correlation function Rq(τ) of the space ﬂuctuations caused by rotation of the polarization plane, simultaneously with radar antenna scanning, takes the following form: Rq (τ) ≈ 0.33(2 + cos 2Ω p τ) .

(5.88)

The corresponding power spectral density Sq(ω) of the space ﬂuctuations caused by rotation of the polarization plane simultaneously with radar antenna scanning has the following form:

Copyright 2005 by CRC Press

1598_book.fm Page 250 Monday, October 11, 2004 1:57 PM

250

Signal and Image Processing in Navigational Systems Sq (ω ) ≅ δ(ω ) + 0.25[δ(ω − 2Ω p ) + δ(ω + 2Ω p )]

(5.89)

and consists of three components:16,17 the main (central) component and two subcomponents offset by ± 2Ωp from the central component. The power of each side component is four times less than the power of the central component. A presence of ± 2Ωp is the consequence of the two-lobe polarization characteristic of scatterers. If a radar antenna with axial symmetry Gaussian directional diagram performs conical scanning, then the normalized correlation function of the target return signal ﬂuctuations, which takes into account the radar antenna scanning and rotation of the polarization plane, can be written under the condition Ωp = Ωsc as the product of Equation (5.75) and Equation (5.88):

R(τ) ≈ 0.33(2 + cos 2Ωsc τ) ⋅ e

− 2π

θ2sc ∆(a2 )

sin 2 0.5 Ωsc τ

.

(5.90)

As before, the power spectral density is the regulated function. The power of harmonics of this power spectral density is determined by PNk ≅ I k (

π θ2sc ∆(a2 )

2 sc (2) a

2 sc (2) a

) + 0.25[I k−2 ( π∆ θ ) + I k+2 ( π∆ θ )] .

(5.91)

Comparing Equation (5.91) with Equation (5.81), one can see an essential increase in the number of harmonics of the power spectral density Sq(ω) of the ﬂuctuations caused by the rotation of the polarization plane of the radar antenna. The second harmonic differs from zero and is equal to 0.25 from the zero harmonic by power due to the rotation of the polarization plane, even if the condition θsc = 0° is satisﬁed. In the case of the cross-polarized signal given by q(ξ , ζ) = sin 2 ξ sin 2 ζ ,

(5.92)

Rq ( ∆ξ , ∆ζ) = Rq ( ∆ξ) = cos 2 ∆ξ ;

(5.93)

Rq (τ) = cos 2Ω p τ .

(5.94)

we can write

The power spectral density with the stationary radar antenna can be written in the following form: Sq (ω ) ≅ δ(ω − 2Ω p ) + δ(ω + 2Ω p )

Copyright 2005 by CRC Press

(5.95)

1598_book.fm Page 251 Monday, October 11, 2004 1:57 PM

Fluctuations Caused by Radar Antenna Scanning

251

and consists only of two subfrequencies without the central component. Under simultaneous conical scanning, i.e., when the condition Ωp = Ωsc is true, the normalized correlation function R(τ) of the ﬂuctuations, which takes into consideration the radar antenna scanning and rotation of the polarization plane, has the following form:

R(τ) = cos 2Ωsc τ ⋅ e

− 2π

θ2sc ∆(a2 )

sin 2 0.5 Ωsc τ

.

(5.96)

The power of harmonics of the power spectral density is determined by the following form: PNk ≅ I k − 2 (

5.5

π θ2sc ∆(a2 )

2 sc (2) a

) + I k+2 ( π∆ θ ) .

(5.97)

Conclusions

If for any navigational system the radar is stationary, the radar antenna scanning is the main source of the slow target return signal ﬂuctuations. In this section we investigate the correlation function and power spectral density of the target return signal ﬂuctuations caused by the moving (rotation) of the radar antenna under scanning of the three-dimensional (space) and two-dimensional (surface) targets. Two of the most universally adopted forms of the radar antenna moving (rotation) are considered: the line scanning used, as a rule, in the detection of targets and the conical scanning used in tracking moving targets. The case of the radar antenna conical scanning with simultaneous rotation of the polarization plane of the radar antenna is also investigated. In the general case, under scanning of the three-dimensional (space) target, the space–time normalized correlation function of ﬂuctuations under conditions discussed in this section is deﬁned by the product of three normalized correlation functions: the normalized correlation function of time ﬂuctuations in the radar range, which is caused by the propagation of the pulsed searching signal; the normalized correlation function of space ﬂuctuations which is caused by the radar antenna moving (rotation); and the normalized correlation function of space ﬂuctuations which is caused by the rotation of the radar antenna polarization plane. In scanning the two-dimensional (surface) target by the pulsed searching signal, the normalized correlation function of the ﬂuctuations can be expressed as the product of the periodic normalized correlation function of the ﬂuctuations in the radar range and the normalized correlation function of the slow ﬂuctuations caused by radar antenna moving (rotation). The Copyright 2005 by CRC Press

1598_book.fm Page 252 Monday, October 11, 2004 1:57 PM

252

Signal and Image Processing in Navigational Systems

power spectral density of the target return signal ﬂuctuations with line scanning of both the three-dimensional (space) target and the two-dimensional (surface) target is the square of the power spectral density of the amplitude of the target return signal, which is the Fourier transform of the individual elementary signal, under covering of all elementary scatterers by the directional diagram during the radar antenna scanning. With line scanning by radar antenna and in the case of the continuous searching signal, the normalized correlation function of the ﬂuctuations in terms of periodic iteration of the same elementary signals with each revolution of the radar antenna is determined by Equation (5.47). The power spectral density is the regulated function given by Equation (5.48). In the case of the pulsed searching signal, the normalized correlation function of the ﬂuctuations caused by the radar antenna line scanning is deﬁned by the product of the normalized correlation function given by Equation (5.47) and the normalized correlation function of the time ﬂuctuations in the radar range, which is determined by Equation (5.2). The power spectral density is the convolution between the power spectral densities given by Equation (3.15) and Equation (5.48). In conical scanning by the radar antenna, the normalized correlation function of the target return signal ﬂuctuations is nonstationary, in the general case, and can be determined by Equation (5.72) that allows us to deﬁne the instantaneous power spectral density of the target return signal ﬂuctuations. If the directional diagram is in axial symmetry, the normalized correlation function of the target return signal ﬂuctuations caused by conical scanning is stationary and the power spectral density is the regulated function. Under radar antenna conical scanning with simultaneous rotation of the polarization plane, the normalized correlation function and power spectral density of the target return signal ﬂuctuations are determined by Equation (5.88) and Equation (5.89) in the case of the polarization plane rotation with uniform angular velocity. The power spectral density of the target return signal ﬂuctuations consists only of three components: the main component and two subcomponents. In the case of the axial symmetry Gaussian directional diagram, the normalized correlation function and power spectral density of the ﬂuctuations caused by conical scanning and simultaneous rotation of the polarization plane of the radar antenna are determined by Equation (5.90) and Equation (5.91), respectively. In this case, there is an essential increase in the number of harmonics of the power spectral density of the target return signal ﬂuctuations.

References 1. Farina, A., Antenna-Based Signal Processing Techniques for Radar Systems, Artech House, Norwood, MA, 1992. 2. Schleher, D., MTI and Pulsed Doppler Radar, Artech House, Norwood, MA, 1991.

Copyright 2005 by CRC Press

1598_book.fm Page 253 Monday, October 11, 2004 1:57 PM

Fluctuations Caused by Radar Antenna Scanning

253

3. Compton, R., Adaptive Antennas, Prentice Hall, Englewood Cliffs, NJ, 1988. 4. Johnson, D. and Dudgeon, D., Array Signal Processing: Concepts and Techniques, Prentice Hall, Englewood Cliffs, NJ, 1993. 5. Hudson, J., Adaptive Array Principles, Peter Peregrinus, London, 1981. 6. Feldman, Yu and Reznikov, V. Reduction of reﬂections under scanning the Earth surface, Problems in Radio Electronics, Vol. OT, No. 1, 1968, pp. 55–66 (in Russian). 7. Rappaport, T., Wireless Communications: Principle and Practice, Prentice Hall, Englewood Cliffs, NJ, 1996. 8. Jablou, N., Adaptive beamforming with generalized side-lobe canceller in the presence of array imperfections, IEEE Trans., Vol. AP-34, No. 8, 1986, pp. 996–1012. 9. Buckley, K. and Grifﬁths, L., An adaptive generalized side-lobe canceller with derivative constraints, IEEE Trans., Vol. AP-34, No. 3, 1986, pp. 311–319. 10. Arnold, H., Cox, D., and Murray, R., Macroscopic diversity performance measured in the 800-MHz portable radio communications environment, IEEE Trans., Vol. AP-36, No. 2, 1988, pp. 277–280. 11. Graziano, V., Propagation correlation at 900 MHz, IEEE Trans., Vol. VT-27, No. 5, 1978, pp. 182–189. 12. Monzingo, R. and Miller, T., Introduction to Adaptive Arrays, John Wiley & Sons, New York, 1980. 13. Feldman, Yu, Gidaspov, Yu, and Gomzin, V., Moving Target Tracking, Soviet Radio, Moscow, 1978 (in Russian). 14. Blackman, S. and Popoli, R., Design and Analysis of Modern Tracking Systems, Artech House, Norwood, MA, 1999. 15. Bogomolov, A., Foundations of Radar, Soviet Radio, Moscow, 1954 (in Russian). 16. Appelbaum, S. and Chapman, D., Adaptive arrays with main beam constraints, IEEE Trans., Vol. AP-24, No. 9, 1976, pp. 650–662. 17. Li, Q., Rothwell, E., Chen, K., and Nyquist, D., Scattering center analysis of radar targets using ﬁtting scheme and genetic algorithm, IEEE Trans., Vol. AP44, No. 1, 1996, pp. 198–207.

Copyright 2005 by CRC Press

1598_C06.fm Page 255 Monday, October 11, 2004 2:59 PM

6 Fluctuations Caused by the Moving Radar with Simultaneous Radar Antenna Scanning

6.1

General Statements

Target return signal ﬂuctuations caused by the moving radar with simultaneous radar antenna scanning are of prime interest. These ﬂuctuations are encountered in navigational systems, for example, when the target is searched against the background of clouds moving under stimulus of the wind.1,2 Fluctuations caused by the moving radar with simultaneous radar antenna scanning are a nonstationary stochastic process.3 This nonstationary stochastic process possesses a set of peculiarities that cannot be investigated by individually studying the ﬂuctuations caused only by the moving radar or those caused only by radar antenna scanning. Problems caused by the moving radar with simultaneous radar antenna scanning are numerous. In this chapter, we study only the most important problems: moving radar with simultaneous radar antenna line scanning and moving radar with simultaneous radar antenna conical scanning.

6.1.1

The Correlation Function in the Scanning of the Three-Dimensional (Space) Target

In the general case, with the pulsed searching signal, the correlation function of the target return signal ﬂuctuations caused by the moving radar with simultaneous radar antenna scanning can be determined by Equation (2.93)–Equation (2.95). The intraperiod and interperiod target return signal ﬂuctuations can be separated because the total normalized correlation function of the target return signal ﬂuctuations is deﬁned by the product of the normalized correlation functions of the intraperiod and interperiod ﬂuctuations. However, because the shifts ∆ϕ, ∆ψ, and ∆ρ are functions of the angles β0 and γ0 — see Equation (2.84)–Equation (2.87) and Equation (2.92), where ∆ρ = ∆ cosβ 0 cos γ 0

(6.1) 255

Copyright 2005 by CRC Press

1598_C06.fm Page 256 Monday, October 11, 2004 2:59 PM

256

Signal and Image Processing in Navigational Systems

is the radial displacement along the radar antenna directional diagram axis — and the angles β0 and γ0 , in turn, are functions of time during the radar antenna scanning, the ﬂuctuations become a nonstationary stochastic process. The character of the nonstationary state of the normalized correlation functions given by Equation (2.94) and Equation (2.95) is different. Dependence on the angles β0(t) and γ0(t) in the normalized correlation function of the ﬂuctuations in the radar range, which is determined by Equation (2.95), manifests itself by the radial displacement ∆ρ0 along the directional diagram axis [see Equation (6.1)]. As the radar moves uniformly with the velocity V, this dependence acts on the coefﬁcient of scale µ given by Equation (3.11) and, as a result, the coefﬁcient µ becomes a function of the time µ = 1 + 2 Vr c −1 = 1 + 2 Vc −1 cosβ 0 ( t ) cos γ 0 ( t ).

(6.2)

This means that the comb structure of the normalized correlation function Rp(t) (see Figure 3.1a) is kept unchanged, but the scale with respect to the variable τ depends on the time t. The instantaneous power spectral density Sp(ω) of the ﬂuctuations in the radar range is the same as in the case of the stationary radar antenna, i.e., it is a regulated function (see Figure 3.1b). However, the scale of Sp(ω) with respect to the axis of frequency is a function of the time. As a rule, we can neglect these changes. The main effect is exhibited by the characteristics of slow (interperiod) ﬂuctuations. First, with the moving radar and stationary radar antenna, the shape, bandwidth, and average frequency of the power spectral density depends on the radar antenna orientation, i.e., the angles β0 and γ0. Second, under radar antenna scanning, the variation of the angles β0 and γ0 shows the regular deformation of the power spectral density, i.e., the nonstationary state. For this reason, in this chapter, the slow ﬂuctuations are the main subject of study. The instantaneous normalized correlation function, which is determined by Equation (2.94), can be written in the following symmetric form: Rg ( ∆, ∆β 0 , ∆γ 0 ) = Rmov ,sc ( ∆, ∆β 0 , ∆γ 0 ) = N

∫∫ g(ϕ − 0.5∆ϕ, ψ − 0.5∆ψ)

× g(ϕ − 0.5 ∆ϕ , ψ − 0.5 ∆ψ ) ⋅ e

4 jπ

∆ρ( ϕ , ψ ) λ

dϕ dψ . (6.3)

The interperiod ﬂuctuations with the pulsed searching signal and the slow ﬂuctuations with the simple harmonic searching signal, in regard to the longrange area, are the same under scanning of the three-dimensional (space) target. The formula in Equation (6.2) contains solutions for both: in the case when the radar is moving and when the radar antenna scanning is absent, i.e., when the radar antenna is stationary and the condition ∆ϕ = ∆ψ = 0° is Copyright 2005 by CRC Press

1598_C06.fm Page 257 Monday, October 11, 2004 2:59 PM

Fluctuations Caused by Moving Radar

257

satisﬁed, and when the radar is stationary and the condition ∆ρ = 0 is true, i.e., there is radar antenna scanning. It can easily be seen that, in the general case, we cannot express the correlation function of the ﬂuctuations caused by the moving radar with simultaneous radar antenna scanning as the product (or another combination) of the correlation function caused by the moving radar and that caused by radar antenna scanning. However, there is at least one exception if the directional diagram is Gaussian.4 There, the total normalized correlation function of the ﬂuctuations caused by the moving radar with simultaneous radar antenna scanning can be written in the following form: Rmov ,sc ( ∆ , ∆β 0 , ∆γ 0 ) = Rmov ( ∆) ⋅ Rsc (∆β 0 , ∆γ 0 ) ,

(6.4)

where

Rmov ( ∆) = N

∫∫ e

− 2π

(ϕ

2

∆(h2 )

+

ψ2 ∆(v2 )

) + 4 jπ ∆ρ(λϕ ,ψ )

dϕ dψ

(6.5)

is the normalized correlation function caused only by the moving radar. Here, ∆ρ[β0(t), γ0(t)] is a function of time;

Rsc ( ∆β 0 , ∆γ 0 ) = e

−0.5 π

2

( ϕ( 2 ) + ∆

h

ψ2 ∆(v2 )

) (6.6)

is the normalized correlation function caused only by radar antenna scanning. Naturally, the total power spectral density of the ﬂuctuations is the convolution between the power spectral density caused only by the moving radar and that caused only by radar antenna scanning. If the directional diagram is not Gaussian, but the variables ϕ and ψ are separable in the function g(ϕ, ψ), then, for a linear approximation of the radar displacement ∆ρ0 along the directional diagram axis, which is determined by Equation (3.47), based on Equation (6.3), we can write h v Rmov ,sc ( ∆, ∆β 0 , ∆γ 0 ) = Rmov , sc ( ∆, ∆β 0 ) ⋅ Rmov , sc ( ∆, ∆γ 0 ) ⋅ e

− 4 jπ

∆ρ0 λ

,

(6.7)

dϕ ;

(6.8)

dψ ;

(6.9)

where

∫

h Rmov gh (ϕ − 0.5∆ϕ) ⋅ gh (ϕ + 0.5∆ϕ) ⋅ e , sc ( ∆, ∆β 0 ) = N

∫

v Rmov gv (ψ − 0.5∆ψ ) ⋅ gv (ψ + 0.5∆ψ ) ⋅ e , sc ( ∆, ∆γ 0 ) = N

Copyright 2005 by CRC Press

− 4jπϕ

− 4jπψ

∆ρh λ

∆ρv λ

1598_C06.fm Page 258 Monday, October 11, 2004 2:59 PM

258

Signal and Image Processing in Navigational Systems

∆ρ0 is given by Equation (6.1); ∆ρh = ∆ ⋅ sin β 0 ;

6.1.2

and

∆ρv = ∆ ⋅ cos γ 0 sin γ 0 .

(6.10)

The Correlation Function in the Scanning of the Two-Dimensional (Surface) Target

In scanning the two-dimensional (surface) target, the correlation function of the target return signal ﬂuctuations [see Equation (2.122)] is seen as more complex.5,6 With the continuous searching signal, we need to consider that the target return signal power depends on the aspect angle and radar range. With the pulsed searching signal, we need to consider that the interperiod and intraperiod ﬂuctuations are not separable (see Chapter 4). If and only if the directional diagram is Gaussian can we deﬁne the normalized correlation function of the ﬂuctuations caused by radar antenna scanning, which is given by Equation (6.6), as the individual cofactor using the deﬁnition of the total correlation function [see Equation (2.122)]. The second cofactor is the normalized correlation function of ﬂuctuations caused by the moving radar, which has been discussed in Chapter 4. If the variables ϕ and ψ are separable in the directional diagram g(ϕ, ψ), then, with the pulsed searching signal, the total correlation function can be expressed as the product of the azimuth-normalized and the aspect-angle normalized correlation functions [see Equation (2.133)–Equation (2.135)]. Under radar antenna scanning, the main changes are seen as arising in the interperiod ﬂuctuations, as in the case of scanning the three-dimensional (space) target. Intraperiod ﬂuctuations have the same character, but only the coefﬁcient of scale µ given by Equation (6.2) becomes a function of β0(t) and γ0(t). In the representation of the pulsed searching signal with short duration, i.e., when the condition ∆p << ∆v is satisﬁed in the form of the delta function assuming τ = nTp′, and using Equation (2.142) and Equation (5.22), based on Equation (2.122), we can write the normalized correlation function Rmov,sc(∆, ∆β0, ∆γ0) of ﬂuctuations caused by the moving radar with simultaneous radar antenna scanning in the following form:

∫

Rmov ,sc ( ∆, ∆β 0 , ∆γ 0 ) = N g(ϕ − 0.5∆ϕ , ψ * − 0.5∆ψ ) × g(ϕ + 0.5 ∆ϕ , ψ * + 0.5 ∆ψ ) ⋅ e

4 jπ

∆ρ( ϕ , ψ ) * λ

(6.11) dϕ ,

where ∆ρ(ϕ , ψ * ) = ∆ ⋅ cos[β 0 (t) +

Copyright 2005 by CRC Press

ϕ cos γ *

] cos γ * ,

(6.12)

1598_C06.fm Page 259 Monday, October 11, 2004 2:59 PM

Fluctuations Caused by Moving Radar

259

γ* = ψ* + γ0 .

(6.13)

The normalized correlation function Rmov,sc(∆, ∆β0, ∆γ0), given by Equation (6.11), is similar to the instantaneous normalized correlation function Rg(∆, ∆β0, ∆γ0) of slow ﬂuctuations given by Equation (6.3). The difference is that an integration is carried out only with respect to the variable ϕ, not the two variables ϕ and ψ. In the general case, Rmov,sc(∆, ∆β0, ∆γ0) is not divided into the normalized correlation function Rmov(∆) of ﬂuctuations caused only by the moving radar and the normalized correlation function Rsc(∆β0, ∆γ0) of ﬂuctuations caused only by radar antenna scanning. In the case of the Gaussian directional diagram, the formula in Equation (6.4) is true, where Rmov(∆) can be written in the following form: ϕ2

∫

Rmov ( ∆) = N e

− 2π

ϕ2 ∆(h2 )

+ 4jπ

∆ρ( ϕ , ψ ∗ ) λ

dϕ .

(6.14)

ϕ1

In this case, Rsc(∆β0, ∆γ0), caused only by radar antenna scanning is completely determined by Equation (6.6). In the case of the Gaussian directional diagram, and if variables ϕ and ψ are separable in the function g(ϕ, ψ), Equation (6.7) and Equation (6.8) are true for a linear approximation of the parameter ∆ρ. Therefore, Equation (6.9) takes the following form:

∫

v gv (ψ * − 0.5∆ψ ) ⋅ gv (ψ * + 0.5∆ψ ) ⋅ e Rmov , sc ( ∆, ∆γ 0 ) = N

− 4jπ ψ *

∆ρv λ

dψ , (6.15)

where ∆ρv = ∆ cos β 0 sin γ * .

6.2 6.2.1

(6.16)

The Moving Radar with Simultaneous Radar Antenna Line Scanning Scanning of the Three-Dimensional (Space) Target: The Gaussian Directional Diagram

Let us assume that the radar moves uniformly and linearly with the velocity V. The radar antenna rotates about its own vertical axis, i.e., the condition γ0 = const is satisﬁed, with the angular velocity Ωsc(t) making the line (circular or segment) scanning the three-dimensional (space) target. First, we consider Copyright 2005 by CRC Press

1598_C06.fm Page 260 Monday, October 11, 2004 2:59 PM

260

Signal and Image Processing in Navigational Systems

the uniform line circular scanning by radar antenna with the constant angular velocity Ωsc = const. In this case, the formula in Equation (6.4) is true. The normalized correlation function Rsc(∆β0, ∆γ0) of ﬂuctuations caused only by radar antenna scanning is Gaussian [see Equation (5.34)]. The normalized correlation function Rmov(∆) of ﬂuctuations caused only by the moving radar depends on orientation of the radar antenna with respect to the vector of velocity V of the moving radar and, in the case of the high-deﬂected radar antenna, is also Gaussian [see Equation (3.79)]. Because of this, the total normalized correlation function Rmov,sc(∆, ∆β0, ∆γ0) of ﬂuctuations caused by the moving radar with simultaneous radar antenna scanning, given by Equation (6.4), can be written in the following form: 2

2

Rmov ,sc (t , τ) = e − π ( ∆Fmov + ∆Fsc ) + jΩ0τ ,

(6.17)

where ∆Fmov is the effective bandwidth of the power spectral density of the ﬂuctuations caused only by the moving radar [see Equation (3.81)], in which β0 = Ωsct is a function of time, ∆Fsc is the effective bandwidth of the power spectral density of the ﬂuctuations caused only by radar antenna scanning [see Equation (5.38)], Ω0 is the averaged Doppler frequency given by Equation (3.48) and Equation (3.49) and is also a function of β0(t). The instantaneous power spectral density of ﬂuctuations caused by the moving radar with simultaneous radar antenna scanning is Gaussian with the effective bandwidth determined in the following form: 2 ∆F ( t ) = ∆Fmov ( t ) + ∆Fsc2 .

(6.18)

More rigorously speaking, to deﬁne the resulting normalized correlation function Rmov,sc(t, τ) of ﬂuctuations caused by the moving radar with simultaneous radar antenna scanning, we must use the normalized correlation function Rsch (τ , nTsc ) of ﬂuctuations caused only by radar antenna scanning, which is determined by Equation (5.47), instead of the normalized correlation function Rsc(τ) given by Equation (5.34). To deﬁne the total instantaneous power spectral density of ﬂuctuations caused by the moving radar with simultaneous radar antenna scanning, we must carry out the convolution between the continuous power spectral density Smov(ω, t) of ﬂuctuations caused only by the moving radar, which is determined by Equation (3.80), and the regulated power spectral density Ssc(ω) of ﬂuctuations caused only by radar antenna scanning, which is given by Equation (5.49). It is not difﬁcult to show that when the effective bandwidth of Smov(ω, t) of ﬂuctuations caused only by the moving radar is signiﬁcantly greater than the frequency of radar antenna scanning, i.e., the condition ∆Fmov >> Fsc or τ mov << Tsc is satisﬁed, the result is the same. c However, the effective bandwidth ∆Fsc of the power spectral density Ssc(ω) of ﬂuctuations caused only by radar antenna line circular scanning can Copyright 2005 by CRC Press

1598_C06.fm Page 261 Monday, October 11, 2004 2:59 PM

Fluctuations Caused by Moving Radar

261

change substantially in value during the radar antenna line circular scanning, especially if the aspect angle γ0 of the radar antenna is not so high in value.7 Under the condition β0 ≈ 0°, the effective bandwidth of the moving radar ∆Fmov of the power spectral density Smov(ω, t) is very low in value and can be less than the frequency of the radar antenna scanning Fsc. Moreover, the lowdeﬂected radar antenna, Smov(ω, t), caused only by the moving radar, is the more complex function, which is not Gaussian (see Section 3.3). In particular, under the condition β0 = 0° and if the radar antenna has axial symmetry, Smov(ω, t) takes the exponential form with the effective bandwidth ∆Fmov that is very low in value. With the Gaussian directional diagram, the total normalized correlation function of the ﬂuctuations Rmov,sc(∆, ∆β0, ∆γ0) caused by the moving radar with simultaneous radar antenna scanning can be expressed as the product of the normalized correlation functions Rmov(∆) of ﬂuctuations caused only by the moving radar and Rsc(∆β0, ∆γ0) caused only by radar antenna scanning. This is despite the fact that with some positions of the radar antenna, the power spectral density Smov(ω, t) of ﬂuctuations caused only by the moving radar does not obey the Gaussian law. Because of this, for all cases, the resulting power spectral density Smov,sc(ω, t) of ﬂuctuations caused by the moving radar with simultaneous radar antenna scanning is deﬁned by the convolution between Smov(ω, t) and the regulated power spectral density Ssc(ω) of ﬂuctuations caused only by radar antenna scanning, which is given by Equation (5.49) and can be written in the following form: ∞

Smov ,sc (ω , t) =

∑S

mov

(ω − kΩsc ) ⋅ e

−π

2 k 2 Ωsc 2 ∆Ωmov

.

(6.19)

k=0

The formula in Equation (6.19) is true for any relationships between ∆Ωmov = 2π∆Fmov and Ωsc. The power spectral density Smov(ω, t) of ﬂuctuations caused by the moving radar with simultaneous radar antenna scanning is the sum of the power spectral densities Smov(ω, t) of ﬂuctuations caused only by the moving radar, which are shifted to the frequencies kΩsc , and multiplied by the coefﬁcients Ssc(kΩsc). Under the condition ∆Ωmov >> Ωsc , we obtain the integral convolution of the already discussed power spectral densities Smov(ω, t) and Ssc(ω) and the already discussed result (see Figure 6.1). Under the condition ∆Ωmov < Ωsc , the power spectral density Smov(ω, t) has a comb structure. In doing so, the shape of the teeth of the comb power spectral density Smov(ω, t) can be changed from the Gaussian form — the radar antenna is high deﬂected and the velocity of the moving radar is low in value (see Figure 6.2) — to the exponential form — the radar antenna is not deﬂected (see Figure 6.3). Thus, during the radar antenna line scanning, the instantaneous power spectral density Smov,sc(ω, t) is deformed continuously taking various shapes in accordance with the position and width of the directional diagram, velocity

Copyright 2005 by CRC Press

1598_C06.fm Page 262 Monday, October 11, 2004 2:59 PM

262

Signal and Image Processing in Navigational Systems

Smov (ω)

ω (a) Ssc (ω)

ω (b) Smov (ω) * Ssc (ω)

ω (c) FIGURE 6.1 Moving radar with simultaneous radar antenna scanning, ∆Fmov >> Fsc . The power spectral density of interperiod ﬂuctuations of the target return signal caused by (a) only moving radar, (b) only radar antenna scanning, and (c) moving radar with simultaneous radar antenna scanning.

of the moving radar, and frequency of radar antenna line scanning. The Doppler shift in the frequency of the partial power spectral densities Smov(ω, t) with regard to the frequencies kΩsc is also continuously varied within the limits of the interval [–Ωmax cos γ0 , Ωmax cos γ0]. With radar antenna segment hunting, the instantaneous angular velocity Ωsc becomes a function of time but Equation (6.4) is kept true. The normalized

Copyright 2005 by CRC Press

1598_C06.fm Page 263 Monday, October 11, 2004 2:59 PM

Fluctuations Caused by Moving Radar

263

Smov (ω)

ω (a) Ssc (ω)

ω (b) Smov (ω) * Ssc (ω)

ω (c) FIGURE 6.2 Moving radar with simultaneous radar antenna scanning. Radar antenna is high deﬂected. Velocity of the moving radar is low by value, ∆Fmov << Fsc . Power spectral density of interperiod ﬂuctuations of the target return signal caused by (a) only moving radar, (b) only radar antenna scanning, and (c) moving radar with simultaneous radar antenna scanning.

correlation function Rsc(∆β0 , ∆γ0) of the ﬂuctuations caused only by the radar antenna scanning becomes more complex. If the condition ∆a << βm (see Section 5.2.3) is satisﬁed, so that we can neglect a variation of the angular velocity during the time of the radar antenna rotation by the angle equal to the directional diagram width — this is called quasi-stationary radar antenna scanning Copyright 2005 by CRC Press

1598_C06.fm Page 264 Monday, October 11, 2004 2:59 PM

264

Signal and Image Processing in Navigational Systems

Smov (ω)

ω (a) Ssc (ω)

ω (b) Smov (ω) * Ssc (ω)

ω (c) FIGURE 6.3 Moving radar with simultaneous radar antenna scanning. Radar antenna is not deﬂected, ∆Fmov << Fsc . The power spectral density of interperiod ﬂuctuations of the target return signal caused by (a) only moving radar, (b) only radar antenna scanning, and (c) moving radar with simultaneous radar antenna scanning.

— Equation (6.17) and Equation (6.18) are true. In this case, the effective bandwidth ∆Fsc of the power spectral density Ssc(ω) of ﬂuctuations caused only by radar antenna scanning becomes a function of time, as the effective bandwidth ∆Fmov of the power spectral density Smov(ω, t) of ﬂuctuations caused only by the moving radar. If the radar antenna segment hunting is rapid, Rsc(∆β0, ∆γ0) is

Copyright 2005 by CRC Press

1598_C06.fm Page 265 Monday, October 11, 2004 2:59 PM

Fluctuations Caused by Moving Radar

265

given by Equation (5.55). Smov,sc(ω, t) of ﬂuctuations caused by the moving radar with simultaneous radar antenna scanning is deﬁned by the corresponding convolution between the power spectral densities Smov(ω, t) and Ssc(ω). 6.2.2

Scanning of the Two-Dimensional (Surface) Target: The Gaussian Directional Diagram

Consider the normalized correlation function Rmov,sc(∆, ∆β0, ∆γ0) of ﬂuctuations of the target return signal caused by the moving radar with simultaneous radar antenna scanning, which is given by Equation (6.11). Using the Gaussian directional diagram, we obtain the product of the two normalized correlation functions: Rsc(∆β0, ∆γ0) of ﬂuctuations caused only by radar antenna scanning, which is determined by Equation (6.6), and Rmov(∆) of the Doppler ﬂuctuations caused by the difference in the radial velocities in the azimuth plane, which is given by Equation (6.14). Because with the highdeﬂected radar antenna, both normalized correlation functions Rmov(∆) and Rsc(∆β0, ∆γ0) obey Gaussian law, the formulae in Section 6.2.1 are true with the only difference that with the pulsed searching signal, the effective bandwidth ∆Fmov of the power spectral density Smov(ω, t) is given by Equation (4.141) instead of Equation (3.81). Also, the effective bandwidth ∆Fsc of the power spectral density Ssc(ω) is given by Equation (5.38), in which the variable γ0 is replaced with the variable γ*. In the case of the continuous searching signal, the formulae obtained in Section 6.2.1 can be used without any changes. In the case of the low-deﬂected radar antenna, we can use the same statements and conclusions as in Section 6.2.1. If it is necessary to obtain rigorous formulae, we must use the results discussed in Section 4.6 to obtain the total power spectral density Smov,sc(ω, t) of the ﬂuctuations caused by the moving radar with simultaneous radar antenna scanning. The main differences rise in the determination of the Fourier transform for the total normalized correlation function Rmov,sc(∆, ∆β0 , ∆γ0). Rmov,sc(∆, ∆β0, ∆γ0) consists of three cofactors: the normalized correlation function Rsch (τ , nTsc ) of the ﬂuctuations caused only by radar antenna scanning, which is given by Equation (5.47); the aspect-angle normalized correlation function Rγ(t, τ) of the interperiod ﬂuctuations caused by the glancing radar range, which is given by Equation (4.67); and the azimuth-normalized correlation function Rβ(τ) of the Doppler ﬂuctuations caused by various values of the azimuth angles within the horizontal-coverage directional diagram width. However, these differences are the same as determined for the power spectral density Smov(ω, t) of the ﬂuctuations caused only by the moving radar in Chapter 4.

6.2.3

Short-Range Area: The Gaussian Directional Diagram

Using an example of scanning the underlying surface,8,9 we consider the short-range area. We need to add the angle shift ∆ϕsc given by Equation (2.84) Copyright 2005 by CRC Press

1598_C06.fm Page 266 Monday, October 11, 2004 2:59 PM

266

Signal and Image Processing in Navigational Systems

and Equation (2.85) caused by radar antenna scanning (naturally, we must replace the variable γ0 with the variable γ*) to the previous angle shift ∆ϕmov given by Equation (2.86) and Equation (2.87) caused by the moving radar. When the radar moves uniformly and the radar antenna rotates uniformly, we can write ∆ϕ = ∆ϕ mov + ∆ϕ sc = [Vρ −∗1 sin β 0 cos γ * + Ω sc cos γ * ] τ.

(6.20)

Based on Equation (6.20) we can obtain that the total normalized correlation function Rmov,sc(∆, ∆β0 , ∆γ0) is Gaussian in the case of the Gaussian directional diagram [see Equation (6.17)]. The effective bandwidth ∆F of the total power spectral density Smov,sc(ω, t) of ﬂuctuations caused by the moving radar with simultaneous radar antenna scanning can be determined in the following form: 2 2 ∆F = ∆Fmov + ∆F˜ sc2 = ∆Fmov + ( ∆Fsc + ∆F˜ mov ) 2

(6.21)

instead of Equation (6.18), where ∆Fmov = 2 Vλ−1∆ h sin β 0

(6.22)

is the effective bandwidth of the power spectral density Sβ(ω) of the Doppler ﬂuctuations caused by the difference in the Doppler frequencies in the azimuth plane; ∆F˜sc = ∆Fsc + ∆F˜mov =

Ωsc 2 ∆h

⋅ cos γ * +

V 2 ρ *∆ h

⋅ sin β 0 cos γ *

(6.23)

is the effective bandwidth of the power spectral density Sγ(ω) of ﬂuctuations caused by the angle displacement of scatterers within the directional diagram width due to radar antenna scanning (the term ∆Fsc) and the moving radar (the term ∆F˜ mov ). Because Ωϕ = Vρ −* 1 sin β 0

(6.24)

is the angular velocity of displacement of scatterers due to the moving radar, we can write ∆F˜sc =

Copyright 2005 by CRC Press

Ωsc +Ωϕ 2 ∆h

⋅ cos γ * .

(6.25)

1598_C06.fm Page 267 Monday, October 11, 2004 2:59 PM

Fluctuations Caused by Moving Radar

267

Here, the following fact is of interest. The angular velocity Ωϕ changes its sign while changing the sign of the azimuth deviation β0. In other words, taking into account the angle shift ∆ϕmov in the short-range area leads to the extension of the effective bandwidth ∆F (the condition β0 > 0 is satisﬁed) and to the contraction of the effective bandwidth ∆F (the condition β0 < 0 is true) of the total power spectral density Smov,sc(ω, t) of the ﬂuctuations caused by the moving radar with simultaneous radar antenna scanning. Thus, the effective bandwidth ∆F of the total power spectral density Smov,sc(ω, t) of ﬂuctuations in the quadrants I and II differs from the effective bandwidth ∆F of the total power spectral density Smov,sc(ω, t) of ﬂuctuations caused by the moving radar with simultaneous radar antenna scanning in the quadrants IV and III, respectively, under the same absolute deviation of the radar antenna beam in the azimuth plane — the phenomenon of asymmetry (see Figure 6.4 and Figure 6.5). This phenomenon can be easily explained based on the physical meaning. The relative displacements of two scatterers spaced by the distance ρ from the radar under the angles ±β with respect to the direction of the moving radar are shown in Figure 6.4. These displacements lead us to the angle shifts ∆ϕmov with various signs depending on the sign of the angle β. At the same time, the radar antenna scanning gives rise to the angle shifts ∆ϕsc with respect to the axis of the directional diagram. The sign of the angle shifts ∆ϕsc is independent of the sign of the angle β and depends only on the direction of the radar antenna scanning. If the radar antenna rotates clockwise, the angle shift ∆ϕmov is subtracted from the angle shift ∆ϕsc in quadrant I and is added to the angle shift ∆ϕsc in quadrant IV. At the surface points with the coordinates satisfying the equality Ωϕ = Ωsc , the angle shifts ∆ϕmov and ∆ϕsc compensate each other, which leads to the radar antenna scanning stopping from the viewpoint of the target return signal ﬂuctuations.10,11 The geometric locus of these surface points forms a circle shown in Figure 6.5 by the solid line. The circle spaced symmetrically in the quadrants III and IV is shown in Figure 6.5 by the dotted line. This circle corresponds to the double effective bandwidth of the power spectral density of the ﬂuctuations caused by the angle displacements of scatterers in comparison with the effective bandwidth ∆F˜ of the power spectral densc

sity Ssc(ω) of the ﬂuctuations caused only by radar antenna scanning.

6.2.4

The Sinc2-Directional Diagram

Let us assume, as before, that V, Ωsc , and γ0 are the constant values. If the radar antenna searches the three-dimensional (space) target, the total normalized correlation function Rmov,sc(∆, ∆β0 , ∆γ0) of ﬂuctuations of the target return signal caused by the moving radar with simultaneous radar antenna scanning can be given by Equation (6.7) — Equation (6.9). In doing so, we must assume that h ∆ψ = 0° in Equation (6.9). The normalized correlation function Rmov (∆, ∆β0) , sc is of prime interest because it contains all the information regarding the moving radar with simultaneous radar antenna scanning. Copyright 2005 by CRC Press

1598_C06.fm Page 268 Monday, October 11, 2004 2:59 PM

268

Signal and Image Processing in Navigational Systems ∆ ∆ϕmov

Ωsc

I

V

IV Ωsc ∆ϕmov ∆ FIGURE 6.4 The power spectral density in the short-range area. The sum of velocities under moving radar with simultaneous radar antenna scanning.

I

II

V

III

IV

FIGURE 6.5 The power spectral density in the short-range area. Geometric locus of compensation (solid line) and the double effective bandwidth (dotted line) of the power spectral density of ﬂuctuations of the target return signal.

Copyright 2005 by CRC Press

1598_C06.fm Page 269 Monday, October 11, 2004 2:59 PM

Fluctuations Caused by Moving Radar

269

h After very complicated mathematical transformations, Rmov , sc ( ∆, ∆β 0 ) can be determined in the following forms: h −3 Rmov , sc (t , τ) = 6 a ⋅ [2 sin ab − sin( a − ab) + a(1 − 0.5b) cos ab − ab cos( a − ab)]

× e jΩ0τ at 0 ≤ b ≤ 0.5 ; (6.26) −3 h ⋅ [sin ab − a(1 − b) cos( a − ab)] ⋅ e Rmov ,sc (t , τ) = 6 a

h Rmov , sc (t , τ) = 0

at

jΩ0τ

0.5 ≤ b ≤ 1; (6.27)

at

b > 1,

(6.28)

where a = 3 π ∆Fsc ⋅| τ |;

(6.29)

b = 0.75∆Fmov ⋅| τ |;

(6.30)

∆Fsc is the effective bandwidth of the power spectral density Ssc(ω) of the ﬂuctuations caused only by radar antenna scanning, which is given by Equation (5.42), at the sinc-directional diagram; ∆Fmov is the effective bandwidth of the power spectral density Smov(ω, t) of the ﬂuctuations caused only by the moving radar, which is given by Equation (3.87), at the sinc-directional diagram (see Figure 6.6). Here the effective bandwidth ∆Fmov is not the total effective bandwidth of the power spectral density Smov(ω, t). It is only that part of the power spectral density Smov(ω, t) that corresponds to the power spectral density of the Doppler ﬂuctuations caused by the difference in the Doppler frequencies in the azimuth plane. The total normalized correlation function Rmov,sc(∆, ∆β0 , ∆γ0) is h deﬁned according to Equation (6.7) by the product of Rmov , sc ( ∆, ∆β 0 ) given by v Equation (6.8) and Rmov ,sc ( ∆ , ∆γ 0 ) given by Equation (6.9). For the case considh ered here, Rmov ,sc ( ∆ , ∆β 0 ) has been determined only by Equation (6.26) — v Equation (6.30) and Rmov ,sc ( ∆ , ∆γ 0 ) is given by Equation (3.88). ∆F

As one can see from Figure 6.6, under the condition q = ∆F sc = 0, i.e., when mov the radar antenna scanning is stopped, the total normalized correlation function Rmov,sc(t, τ) of the ﬂuctuations caused by the moving radar with simultaneous radar antenna scanning is smooth and coincides with the normalized v correlation function Rmov ,sc ( t , τ ) given by Equation (3.88). With an increase in the value of q, we can observe the side lobes, the level of which increases and at q = 2 reaches the maximum (for about 5%). As q → ∞, i.e., the radar is stationary, the number of side lobes increases, but their level decreases h and Rmov,sc(t, τ) becomes a smooth function again and coincides with Rmov ,sc (τ ) given by Equation (5.39). Copyright 2005 by CRC Press

1598_C06.fm Page 270 Monday, October 11, 2004 2:59 PM

270

Signal and Image Processing in Navigational Systems

Rmov,sc (τ) 1.0 1 10−1

2

3

4

4

5 10−2 6

5 10−3

6

5 6

10−4

0

4

∆Fmov τ

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3

FIGURE 6.6 The normalized correlation function of ﬂuctuations of the target return signal caused by the moving radar with simultaneous radar antenna line scanning for the sinc-directional diagram ∆F of radar antenna at various values of q = ∆F sc : (1) q = 0; (2) q = 0.5; (3) q = 1; (4) q = 2; (5) q = 5; mov (6) q = 40.

Under scanning the two-dimensional (surface) target with the pulsed searching signals and radar antenna line scanning to deﬁne the total normalized correlation function Rmov,sc(t, τ) it is necessary, according to Section v 6.1.2, to take the product of Rmov given by Equation (6.26) and Equation ,sc ( t , τ ) (6.30), in which the variable γ0 is replaced with the variable γ*, and Rγ(t, τ) given by Equation (2.135). If we use the short pulsed searching signal, i.e., the duration of the pulsed searching signal is low in value so that the correlation interval given by Equation (4.70) is much more than that of v Rmov ,sc ( t , τ ), given by Equation (6.26)–Equation (6.30), it characterizes completely the interperiod ﬂuctuations caused by the moving radar with simultaneous radar antenna scanning.

6.3 6.3.1

The Moving Radar with Simultaneous Radar Antenna Conical Scanning The Instantaneous Power Spectral Density

Let us assume that the radar moving uniformly with the velocity V and searching the three-dimensional (space) target makes the open conical scanning by radar antenna with the angular velocity Ωsc , and the angle between the directional diagram axis and the equisignal direction is equal to θsc.12 The equisignal

Copyright 2005 by CRC Press

1598_C06.fm Page 271 Monday, October 11, 2004 2:59 PM

Fluctuations Caused by Moving Radar

271

A α0

θsc

C θp

θ0

V O

FIGURE 6.7 The coordinate system under the moving radar with simultaneous radar antenna conical scanning: OC — equisignal direction; OA — the axis of the radar antenna directional diagram.

direction is deﬂected on the arbitrary angle θp from the vector of velocity V (see Figure 6.7). Then, the directional diagram axis at the arbitrary instant of time is set under the angle θ 0 = θ 2p + θ 2sc + 2θ pθ sc ⋅ cosα 0

(6.31)

with respect to the vector of velocity V, where α0 = Ωsct is the angle characterizing the moving position of the radar antenna. The angle α0 is reckoned from a plane passing through the equisignal direction and the vector of velocity V (see Figure 6.7). If the condition θp >> θsc is satisﬁed, we can write θ 0 ≈ θ p + θ sc ⋅ cosα 0 .

(6.32)

If the condition θp = 0° is true, then θp = θsc = const. The shape, bandwidth, and average frequency of the power spectral density Smov(ω, t) of the ﬂuctuations caused only by the moving radar, as was shown in Chapter 3, depend essentially on the angle θ0. In other words, during the radar antenna conical scanning, the power spectral density of the ﬂuctuations Smov(ω, t) caused only by the moving radar is continuously deformed, i.e., Smov(ω, t) becomes a function of time. With the Gaussian directional diagram, the instantaneous normalized correlation function Rmov,sc(ω, t) of the ﬂuctuations caused by the moving radar with simultaneous radar antenna scanning can be determined by Equation (6.4). The total instantaneous power spectral density of ﬂuctuations of Smov,sc(ω, t) caused by the moving radar with simultaneous radar antenna conical scanning is deﬁned by the convolution between Smov(ω, t) caused only by the moving radar and Ssc(ω) caused only by radar antenna conical scanning.

Copyright 2005 by CRC Press

1598_C06.fm Page 272 Monday, October 11, 2004 2:59 PM

272

Signal and Image Processing in Navigational Systems

With the axially symmetric Gaussian directional diagram of the radar antenna, Smov,sc(ω, t) caused by the moving radar with simultaneous radar antenna conical scanning is deﬁned by the convolution between Smov (ω, t) of ﬂuctuations caused only by the moving radar, which is given by Equation (3.80), and Ssc(ω) of ﬂuctuations caused only by radar antenna conical scanning, which is given by Equation (5.81), and can be written in the following form: ∞

Smov ,sc (ω , t) ≅

∑I [ k

k=0

π θ2sc ∆(a2 )

] ⋅ Smov (ω − kΩsct) .

(6.33)

The coefﬁcients of the series Ik(x) with the used values of the parameter

πθ2sc ∆(a2 )

decrease fast, so that in practice the process can be limited only by three terms: k = –1, 0, 1.13 If the effective bandwidth ∆Fmov of the power spectral density Smov(ω, t) of ﬂuctuations caused only by the moving radar is satisﬁed by the condition ∆Fmov < Fsc , where Fsc is the frequency of radar antenna conical scanning, the total instantaneous power spectral density of ﬂuctuations Smov,sc(ω, t) caused by the moving radar with simultaneous radar antenna conical scanning, which is given by Equation (6.33), consists of three individual parts spaced near the frequencies: ω0 + Ω0, ω0 + Ω0 + Ωsc , and ω0 + Ω0 – Ωsc . The shape of these individual parts of Smov,sc(ω, t) depends on a value of the angle θ0 (see Figure 6.8 and Figure 6.9). Under the high-deﬂected equisignal direction, every individual part of Smov,sc(ω, t) has the Gaussian form (see Figure 6.8). Under the nondeﬂected equisignal direction, every individual part of Smov,sc(ω, t) has the shape of the function given by Equation (3.141) (see Figure 6.9). If the condition ∆Fmov > Fsc is satisﬁed, these individual parts of Smov,sc(ω, t) are overlapped. Under the Smov,sc (ω)

ω ω0 + Ω0 − Ωsc

ω0 + Ω0

ω0 + Ω0 + Ωsc

FIGURE 6.8 The instantaneous power spectral density of ﬂuctuations of the target return signal under the moving radar with simultaneous radar antenna conical scanning. Equisignal direction is high deﬂected, and ∆Fmov << Fsc .

Copyright 2005 by CRC Press

1598_C06.fm Page 273 Monday, October 11, 2004 2:59 PM

Fluctuations Caused by Moving Radar

273

Smov,sc (ω)

ω ω0 + Ω0 − Ωsc

ω0 + Ω0 + Ωsc

ω0 + Ω0

FIGURE 6.9 The instantaneous power spectral density of ﬂuctuations of the target return signal under the moving radar with simultaneous radar antenna conical scanning. Equisignal direction is not deﬂected, and ∆Fmov << Fsc .

Smov,sc (ω)

ω ω0 + Ω0 − Ωsc

ω0 + Ω0

ω0 + Ω0 + Ωsc

FIGURE 6.10 The instantaneous power spectral density of ﬂuctuations of the target return signal under the moving radar with simultaneous radar antenna conical scanning. Equisignal direction is high deﬂected, and ∆Fmov >> Fsc .

condition ∆Fmov >> Fsc , the effective bandwidth and shape of Smov,sc(ω, t) are deﬁned by the effective bandwidth and shape of the central part (see Figure 6.10). In practice, the condition ∆Fmov >> Fsc is true when the radar antenna is high deﬂected, and the condition ∆Fmov < Fsc is true with the nondeﬂected radar antenna.14,15

6.3.2

The Averaged Power Spectral Density

In target tracking in navigational systems, the averaged power spectral density S mov,sc (ω , t ) of ﬂuctuations of the target return signal caused by the moving radar with simultaneous radar antenna conical scanning is more important compared to the total instantaneous power spectral density Smov,sc(ω, t) of ﬂuctuations of the target return signal caused by the moving radar with simultaneous radar antenna conical scanning. This is because the target return signal is subjected to only frequency signal processing, not Copyright 2005 by CRC Press

1598_C06.fm Page 274 Monday, October 11, 2004 2:59 PM

274

Signal and Image Processing in Navigational Systems

frequency–time signal processing, as in navigational systems where purpose of the surveillance radar is to pick out the frequency and phase of radar antenna conical scanning.16,17 To obtain the averaged power spectral density S mov,sc (ω , t ) of ﬂuctuations, we need to average the power spectral density Smov[ω, θ0 (t)] of the ﬂuctuations caused only by the moving radar, which is given by Equation (3.80), during the period Tsc of radar antenna scanning for every term of Smov, sc(ω, t), which is given by Equation (6.33). We must bear in mind that the parameter θ0 is given by Equation (6.31). Under the nondeﬂected equisignal direction, i.e., when the condition θp = 0° is satisﬁed, the equality θ0 = θsc is true and the parameter θ0 does not depend on time. The target return signal is the stationary stochastic process and there is no need to average it. Smov , sc (ω , t), which is given by Equation (6.33), coincides with the averaged Smov , sc (ω , t). Under the high-deﬂected equisignal direction, this problem can be solved only approximately.18 Let us consider the total instantaneous normalized correlation function Rmov,sc(t, τ) of the ﬂuctuations caused by the moving radar with simultaneous radar antenna conical scanning. For the case considered here, we can write Rmov,sc(t, τ) in the following form [see Equation (3.79) and Equation (3.83)]: Rmov,sc ( t , τ ) = e

2 − π ∆Fmov τ 2 −2δ sc sin 2 ( 0.5Ωsc τ ) + jΩmax cosθ 0 ( t )τ

,

(6.34)

where ∆Fmov = 2 Vλ−1∆ a sin θ 0 ; δ sc =

π θ 2sc ∆(a2)

(6.35)

(6.36)

is the parameter of radar antenna conical scanning; the function θ0(t) is given by Equation (6.32). Thus, Rmov, sc(t, τ) is a function of time, due to the effective bandwidth ∆Fmov of the power spectral density Smov(ω, t) of the ﬂuctuations caused only by the moving radar and the averaged Doppler frequency Ωmax × cos θ0 being functions of the parameter θ0(t). With the high-deﬂected radar antenna, i.e., with the condition θp >> θsc being satisﬁed, changes in the effective bandwidth ∆Fmov of the power spectral density Smov(ω, t) are very low in value, and so can be neglected. We must use only the angle θp in Equation (6.35) instead of the angle θ0. Changes in frequency are very important. Because of this, we can write cosθ 0 ≈ cosθ p − θ sc sin θ p ⋅ cosΩ sc t. Copyright 2005 by CRC Press

(6.37)

1598_C06.fm Page 275 Monday, October 11, 2004 2:59 PM

Fluctuations Caused by Moving Radar

275

Taking into consideration Equation (6.37) and averaging Rmov,sc(t, τ), which is given by Equation (6.34), with respect to the variable t, the averaged normalized correlation function Rmov,sc ( t , τ ) of ﬂuctuations caused by the moving radar with simultaneous radar antenna conical scanning can be written in the following form: Rmov , sc (t , τ) = I 0

[

]

8πδ sc ∆Fmov , θp τ ⋅ e

2 − π ∆Fmov τ 2 − 2 δ sc ⋅sin 2 ( 0.5 Ωsc τ ) + jΩ pτ ,θp

,

(6.38)

where Ω p = Ωmax ⋅ cos θ p

(6.39)

is the Doppler shift in the frequency corresponding to the equisignal direction; and ∆Fmov , θp = 2 V λ −1 ∆ a sin θ p

(6.40)

is the effective bandwidth of the power spectral density Smov(ω, t) of ﬂuctuations caused only by the radar moving when the directional diagram axis is deﬂected by the angle θp . As one can see from Figure 6.11, under the condition Fsc ≤ ∆Fmov,θp , the envelope of the averaged normalized correlation function Rmov , sc (t , τ), which is given by Equation (6.38), is smoothed. Under Fsc > ∆Fmov, θp, the envelope of Rmov , sc (t , τ) is wavy. Therefore, the number of waves is high, and the value of the frequency Fsc of radar antenna conical scanning is also high. The Fourier transform of Rmov , sc (t , τ) which is given by Equation (6.38), is not determined exactly. The computer-calculated averaged power spectral densities Smov , sc (ω , t) of ﬂuctuations caused by the moving radar with simultaneous radar antenna conical scanning are shown in Figure 6.12. Under the condition Fsc ≤ ∆Fmov,θp , Smov , sc (ω , t) is smoothed. Under the condition Fsc > ∆Fmov, θp , Smov , sc (ω , t) have the waves corresponding to the frequency Fsc of radar antenna conical scanning. The averaged normalized correlation function Rmov , sc (t , τ) of ﬂuctuations caused by the moving radar with simultaneous radar antenna conical scanning can be determined in the following form based on results discussed in Feldman et al.:19

Copyright 2005 by CRC Press

1598_C06.fm Page 276 Monday, October 11, 2004 2:59 PM

276

Signal and Image Processing in Navigational Systems

Rmov,sc (τ) 1.0 0.9 0.8 0.7 0.6 1

0.5 3

0.4

2 0.3 0.2 0.1

∆Fmov,θp τ

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 FIGURE 6.11 The envelope of the averaged normalized correlation function of ﬂuctuations of the target return ∆θ signal under the moving radar with simultaneous radar antenna conical scanning, ∆ sc = 4: a F F F (1) ∆Fsc = 0.2; (2) ∆Fsc = 1; (3) ∆Fsc = 5. mov

mov

mov

Smov,sc (ω) 1.0 0.9 0.8 0.7 0.6 0.5 0.4 1 0.3 2

2

0.2

3 ω − ω0 + Ωp

0.1

2π∆Fmov,θp

0

1

2

3

4

5

6

FIGURE 6.12 The envelope of the averaged power spectral density of ﬂuctuations of the target return signal under the moving radar with simultaneous radar antenna conical scanning, ∆∆θsc = 4: (1) ∆FFsc a mov = 0.2; (2) ∆FFsc = 1; (3) ∆FFsc = 5. mov

Copyright 2005 by CRC Press

mov

1598_C06.fm Page 277 Monday, October 11, 2004 2:59 PM

Fluctuations Caused by Moving Radar

Rmov , sc (t , τ) ≈ e

− π ∆F12 τ 2 − 2 δ sc ⋅sin 2 ( 0.5 Ωsc τ ) + jΩ pτ

277

,

(6.41)

where ∆F1 =

e δ sc ⋅ ∆Fmov ,θp . I 0 (δ sc )

(6.42)

The resulting instantaneous power spectral density Smov, sc(ω, t) of ﬂuctuations is deﬁned by the convolution between the Gaussian power spectral density Smov(ω, t) of ﬂuctuations caused only by the moving radar and the power spectral density Ssc(ω) of ﬂuctuations caused only by radar antenna conical scanning, given by Equation (5.81) that corresponds to Figure 6.12. All formulae obtained and discussed in this section, results, and conclusions are true for navigational systems employing the radar with the hidden radar antenna conical scanning. In this case, we must replace the parameter θsc with the parameter 0.5θsc .

6.4

Conclusions

Target return signal ﬂuctuations caused by the moving radar with simultaneous radar antenna scanning are the nonstationary stochastic process, possessing a set of peculiarities that cannot be investigated by individually studying ﬂuctuations caused only by the moving radar or caused only by radar antenna conical scanning. Here, the moving radar with simultaneous radar antenna line scanning and with simultaneous radar antenna conical scanning are discussed. When scanning both the three-dimensional (space) target and the twodimensional (surface) target, in general, the normalized correlation function of the ﬂuctuations caused by the moving radar with simultaneous radar antenna conical scanning cannot be expressed as the product of the normalized correlation function of ﬂuctuations caused only by the moving radar and those caused only by radar antenna scanning. An exception is the case where the directional diagram is Gaussian. For this, the total instantaneous power spectral density of the ﬂuctuations caused by the moving radar with simultaneous radar antenna scanning is deﬁned by the convolution between the power spectral densities of ﬂuctuations caused only by the moving radar and only by radar antenna scanning. Under the moving radar with simultaneous radar antenna uniform line circular scanning, the normalized correlation function of the target return signal ﬂuctuations can be expressed as the product of the normalized correlation function of the ﬂuctuations caused only by the moving radar and that Copyright 2005 by CRC Press

1598_C06.fm Page 278 Monday, October 11, 2004 2:59 PM

278

Signal and Image Processing in Navigational Systems

caused only by radar antenna line scanning. In this case, the normalized correlation function of the ﬂuctuations caused only by the moving radar depends on the orientation of the radar antenna with respect to the velocity vector of the moving radar and obeys the Gaussian law with the highdeﬂected radar antenna. The normalized correlation function of ﬂuctuations by radar antenna line scanning is also Gaussian. The total instantaneous power spectral density of ﬂuctuations caused by the moving radar with simultaneous radar antenna line scanning obeys the Gaussian law. During the radar antenna line scanning, the instantaneous power spectral density of ﬂuctuations caused by the moving radar with simultaneous radar antenna line scanning is deformed, continuously taking various shapes in accordance with the position and width of the directional diagram, velocity of moving radar, and frequency of radar antenna line scanning. In the case of the moving radar with simultaneous radar antenna conical scanning, and when the directional diagram is Gaussian, the instantaneous normalized correlation function of ﬂuctuations is expressed as the product of the normalized correlation functions of ﬂuctuations caused only by the moving radar and those caused only by radar antenna conical scanning. The total instantaneous power spectral density of ﬂuctuations caused by the moving radar with simultaneous radar antenna conical scanning is deﬁned by the convolution between the instantaneous power spectral densities of ﬂuctuations caused only by the moving radar and those caused only by radar antenna conical scanning.

References 1. Blackman, S., Multiple-Target Tracking with Radar Applications, Artech House, Norwood, MA, 1986. 2. Alexander, S., Adaptive Signal Processing: Theory and Applications, Springer-Verlag, New York, 1986. 3. Haykin, S., Communication Systems, 3rd ed., John Wiley & Sons, New York, 1994. 4. Rappaport, T., Wireless Communications: Principles and Practice, Prentice Hall, Englewood Cliffs, NJ, 1996. 5. Godard, D., Self-recovering equalization and carrier tracking in two-dimensional data communication systems, IEEE Trans., Vol. COM-28, No. 11, 1980, pp. 1867–1875. 6. Headrick, J. and Skolnik, M., Over-the-horizon radar in the HF band, in Proceedings of the IEEE, Vol. 6, No. 6, 1974, pp. 664–673. 7. Jao, J., A matched array beamforming technique for low angle radar tracking in multipath, in Proceedings of the IEEE National Radar Conference, 1994, pp. 171–176. 8. Krolik, J. and Andersen, R., Maximum likelihood coordinate registration for over-the-horizon radar, IEEE Trans., Vol. SP-45, No. 4, 1997, pp. 945–959. 9. Papazoglou, M. and Krolik, J., Matched-ﬁeld estimation of aircraft altitude from multiple over-the-horizon radar revisits, IEEE Trans., Vol. SP-47, No. 4, 1999, pp. 966–976. Copyright 2005 by CRC Press

1598_C06.fm Page 279 Monday, October 11, 2004 2:59 PM

Fluctuations Caused by Moving Radar

279

10. Kohno, R. et al., Combination on adaptive array antenna and a canceller of interference for direct-sequence spread-spectrum multiple-access systems, IEEE J. Select. Areas Commun., Vol. 8. No. 5, 1990, pp. 641–649. 11. Widrow, B. and Stearus, S., Adaptive Signal Processing, Prentice Hall, Englewood Cliffs, NJ, 1985. 12. Vaidyanathau, C. and Buckley, K., An adaptive decision feedback equalizer antenna array for multiuser CDMA wireless communications, in Proceedings of the 30th Asilomar Conference on Circuits, Systems, Computers, Paciﬁc Grove, CA, November 1996, pp. 340–344. 13. Machi, O., Adaptive Processing, John Wiley & Sons, New York, 1995. 14. Schmidt, R., Multiple emitter location and signal parameter estimation, IEEE Trans., Vol. AP-34, No. 3, 1986, pp. 276–280. 15. McNamara, L., The Ionosphere: Communications, Surveillance, and Direction Finding, Krieger, Malabar, FL, 1991. 16. Monzingo, R. and Miller, T., Introduction to Adaptive Arrays, John Wiley & Sons, New York, 1980. 17. Benedetto, S. and Biglieri, E., Non-linear equalization of digital satellite channels, IEEE J. Select. Areas Commun., Vol. SAC-1, No. 1, 1983, pp. 57–62. 18. Liu, H. and Zoltowski, M., Blind equalization in antenna array CDMA systems, IEEE Trans., Vol. SP-45, No. 1, 1997, pp. 161–172. 19. Feldman, Yu, Gidaspov, Yu, and Gomzin, V., Moving Target Tracking, Soviet Radio, Moscow, 1978 (in Russian).

Copyright 2005 by CRC Press

1598_book.fm Page 281 Monday, October 11, 2004 1:57 PM

7 Fluctuations Caused by Scatterers Moving under the Stimulus of the Wind

7.1

Deterministic Displacements of Scatterers under the Stimulus of the Layered Wind

Fluctuations of the target return signal caused by displacements of scatterers under the stimulus of the wind are of great interest.1–7 Let us consider the slow ﬂuctuations of the target return signal caused, for example, by a moving cloud of scatterers under the stimulus of the wind, gravity, stream of reactive aircraft, and other sources. As regards the rapid ﬂuctuations of the target return signal in the radar range with the pulsed searching signal, all statements and conclusions made in Section 3.1.2 are true for the case considered there. With the continuous searching signal, the slow ﬂuctuations completely cover the whole spectrum of ﬂuctuation sources. The moving radar can be considered as the particular case of the consistent motion of scatterers with the velocity equal in value to the velocity of the moving radar with the opposite sign. In the general case, the motion of the cloud of scatterers is conveniently split into two components:8,9 the deterministic motion of scatterers with various velocities, which can be deﬁned by the nonstochastic function of coordinates and time (for example, variations in the velocity of the wind as a function of altitude [the layered wind], i.e., the motion of cloud of scatterers as a whole) and the stochastic motion of scatterers with the velocity at random and varied in time, in the general case. Let us consider the ﬂuctuations caused by the deterministic displacements of scatterers, in particular, the simultaneous stimulus of the layered wind and moving radar. In principle, this problem is analogous to the problem of the moving radar only and was investigated in more detail in Chapter 3. We can solve this problem and obtain a solution if at the given law of motion of scatterers under the stimulus of the wind we can deﬁne the radial displacements ∆ρw of scatterers as the function of the angle coordinates β and γ. Obviously, the radial displacements ∆ρw of scatterers must be added to the displacements ∆ρr of scatterers caused by the moving radar.

Copyright 2005 by CRC Press

1598_book.fm Page 282 Monday, October 11, 2004 1:57 PM

282

Signal and Image Processing in Navigational Systems

Let us assume that the velocity Vw of the wind and the velocity of the moving scatterers vary linearly as a function of altitude

Vw = Vw0 +| g|h ,

(7.1)

where Vw0 is the velocity of the wind at the altitude h at the point of cloud of the scatterers; g is the gradient of velocity of the wind; and h is the altitude reckoned from the plane of the moving radar (see Figure 7.1). Let us suppose that the velocity of the wind is directed under the angle βw with respect to the velocity of the moving radar. As the following equality

h = ρ sin γ

(7.2)

is true, where ρ is the distance between the radar and the corresponding pulsed volume, the radial components of the velocity of scatterers can be determined in the following form:

Vwr = Vw cos(β − β w )cos γ = (Vw0 +| g|ρ sin γ )cos(β − β w ) cos γ . (7.3) Let us take into consideration the moving radar. The velocity of the moving radar Va relative to the Earth’s surface is equal to the vector sum of the

gh • .•

•

ρ

•h

γ0

• •

β0 βw

Vw

0

Vr

• • • • •

FIGURE 7.1 The power spectral density formation with the layered wind.

Copyright 2005 by CRC Press

1598_book.fm Page 283 Monday, October 11, 2004 1:57 PM

Fluctuations Caused by Moving Scatterers

283

velocity of the moving radar in air Vr and the velocity Vw′ 0 of the wind at the point of the radar in space. If the distance between the radar and the cloud of scatterers, which forms the target return signal, is high in value, then the velocities Vw0 and Vw′0 cannot be equal. The projection of the velocity Va of the moving radar relative to the Earth’s surface in the direction deﬁned by the angles β and γ can be written in the following form:

Var = Vr cos β cos γ + Vw′0 cos(β − β w )cos γ .

(7.4)

The radial component of the velocity of scatterers relative to the radar is determined by

r Vscat = Var − Vwr = {Vr cos β + (∆Vw0 −| g|ρ sin γ )} cos γ ,

(7.5)

∆Vw0 = Vw′0 − Vw0 .

(7.6)

where

Using Equation (2.79), we can deﬁne the angles β and γ as the function of the variables ϕ and ψ. We use the Taylor-series expansion for the radial r component Vscat , limiting to linear terms r Vscat = V0′ − ϕ ⋅ Vh′ − ψ ⋅ Vv′ ,

(7.7)

where

V0′ = Vr cos β 0 cos γ 0 + (∆Vw0 −| g|ρ sin γ 0 ) cos(β 0 − β w ) cos γ 0 ; (7.8) Vh′ = Vr sin β 0 + (∆Vw0 −| g|ρ sin γ 0 ) sin(β 0 − β w ) ;

(7.9)

Vv′ = Vr cos β 0 sin γ 0 + (∆Vw0 sin γ 0 +| g|ρ cos 2 γ 0 ) cos(β 0 − β w ). (7.10) Further investigation can be carried out as well for the high-deﬂected radar antenna using the formulae in Section 3.2, in which the variables Ω0 , Ωh , Ωv must be replaced with the following variables:

Ω′0 = 4 π V0′λ −1 ; Ω′h = 4 π Vh′λ −1 ; and Ω′v = 4 π Vv′λ −1 ;

Copyright 2005 by CRC Press

(7.11)

1598_book.fm Page 284 Monday, October 11, 2004 1:57 PM

284

Signal and Image Processing in Navigational Systems

respectively. If the wind is absent, Equation (7.11) coincides with Equation (3.48)–Equation (3.51). As a result, the power spectral density of the ﬂuctuations is given by Equation (3.70), in which the variables Ω0′ and

A = Ω′h2 + Ω′v2

(7.12)

determined by Equation (3.48)–Equation (3.51) must be replaced with the variables Ω0′ , Ωh′, and Ωv′ given by Equation (7.11). As before, the variable Ω0′ deﬁnes the Doppler shift in the averaged frequency of the power spectral density of the ﬂuctuations. Because the velocity of the wind can have any sign, the presence of the wind leads to both increase and decrease in the averaged frequency. According to Equation (3.70), the power spectral density of the ﬂuctuations coincides in shape with the squared directional diagram by power in the plane passing through the plane ϕ (the horizon, see Figure 2.7) under the angle

κ ′ = arctg

Ω′v Ω′h

.

(7.13)

However, now this plane does not pass through the direction of the moving radar and the directional diagram axis, neither does it occur in the absence of the wind. If the condition in Equation (3.73) is satisﬁed, Equation (3.74) should be used to deﬁne the power spectral density of the ﬂuctuations. In this case, it takes the following form:

S(ω) = gh2 ( ω −ωA0 −Ω′0 ),

(7.14)

where gh (…) is the directional diagram in the plane κ′. The effective bandwidth of the power spectral density of the ﬂuctuations is deﬁned, as in Equation (3.76), by the squared directional diagram width by power in the plane κ′:

∆F = 0.5π −1 A ∆(κ2 ) .

(7.15)

The value of ∆F, as in Equation (3.76), is independent of the parameter λ, but unlike Equation (3.76), it is a complex function of the angles β0 , γ0 , and βw . Now let us assume that the directional diagram is Gaussian. Then, the normalized correlation function and the power spectral density of the ﬂuctuations are given by Equation (3.79) and Equation (3.80), respectively, in which the effective bandwidth is determined by the following form:

Copyright 2005 by CRC Press

1598_book.fm Page 285 Monday, October 11, 2004 1:57 PM

Fluctuations Caused by Moving Scatterers

285

∆F = 2 λ −1{∆(h2 )[Vr sin β 0 + (∆Vw0 −| g|ρ sin γ 0 ) sin(β 0 − β w )] 2 + ∆(v2 )[Vr cos β 0 sin γ 0 + (∆Vw0 sin γ 0 +| g|ρ cos 2 γ 0 ) cos(β 0 − β w )] 2 } 0.5 . (7.16) The formula in Equation (7.16) is the generalization of Equation (3.81) for the case of the proper deterministic motion of scatterers. In doing so, under the conditions ∆Vw = 0 and |g| = 0 — i.e., the wind is absent — 0 Equation (3.81) follows from Equation (7.16). Because in the general case the values of ∆Vw and |g| can have any sign, the presence of the wind 0 leads to both extension and narrowing of the effective bandwidth of the power spectral density of the target return signal ﬂuctuations caused by the moving radar. Dependence of the effective bandwidth ∆F of the power spectral density of the ﬂuctuations on the parameter ρ — the distance between the radar and the target — is a very interesting peculiarity of Equation (7.16). This can be explained by the fact that, in accordance with the parameter ρ, the directional diagram illuminates an area of the cloud of scatterers, which have various velocities as a function of altitude. Let us consider some particular cases.

7.1.1

The Radar Antenna Is Deflected in the Horizontal Plane

In this case, the condition γ0 = 0 is true. Let ∆Vw = 0. Then, 0

Ω′0 = 4 π Vr λ −1 cos β 0 ;

(7.17)

2 ∆F = 2 λ −1 ∆(h2 )Vr2 sin 2 β 0 + ∆ (v2 ) | g|2 ρ2 cos 2 (β 0 − β w ) = ∆Fmov + ∆Fw2 , (7.18) where ∆Fmov is the effective bandwidth of the power spectral density of the target return signal ﬂuctuations caused only by the moving radar, which is given by Equation (3.81); ∆Fw is the effective bandwidth of the power spectral density of ﬂuctuations of the target return signal caused only by the stimulus of the wind (the radar is stationary). Reference to Equation (7.18) shows that if the radar antenna is deﬂected in the azimuth plane and the aspect angle is low in value, the stimulus of the layered wind leads always to extension of the power spectral density of the ﬂuctuations caused by the moving radar.

Copyright 2005 by CRC Press

1598_book.fm Page 286 Monday, October 11, 2004 1:57 PM

286 7.1.2

Signal and Image Processing in Navigational Systems The Radar Antenna Is Deflected in the Vertical Plane

In this case, the condition β0 = 0° is true. Let us also assume that the direction of the wind coincides with the direction of the moving radar, i.e., the conditions ∆Vw = 0 and βw = 0° are satisﬁed. Then we can write 0

Ω′0 = 4πλ −1(Vr −| g|ρ sin γ 0 ) cos γ 0 ;

(7.19)

∆F = 2 ∆ h λ −1|Vr sin γ 0 +| g|ρ cos 2 γ 0|| = ∆Fmov ′ + ∆Fw′|,

(7.20)

where ∆Fmov ′ is the effective bandwidth of the power spectral density of ﬂuctuations caused only by the moving radar; ∆Fw′ is the effective bandwidth of the power spectral density of ﬂuctuations caused only by the stimulus of the wind. If the radar antenna is deﬂected in the vertical plane and the direction of the wind is matched with the direction of the moving radar, the effective bandwidth of the resulting power spectral density of the ﬂuctuations is deﬁned by the algebraic sum of the effective bandwidths of the power spectral density of the ﬂuctuations caused only by the moving radar and that caused only by the stimulus of the wind. In this case, both the extension and the narrowing of the power spectral density are possible because components can be compensated by each other. With some values of the angle γ0 and the distance ρ, the effective bandwidth of the power spectral density of the ﬂuctuations is equal to zero for the considered approximation. Narrowing of the power spectral density can be easily understood if we take into consideration the fact that with an increase in the value of the aspect angle γ, the radial component of the velocity of the moving radar decreases and the radial component of the velocity of the wind increases with the corresponding sign of the gradient of velocity of the wind. Due to this fact, their sum can be approximately constant within the limits of the narrow interval of the aspect angle γ. This phenomenon is illustrated in Figure 7.2, where the segments AB and A′B′ are the projections of the velocityVr on two different directions within the directional diagram. The segments BC and B′C′ are the projections of the velocity| g|h of the wind as functions of altitude. As one can see from Figure 7.2, under deﬁnite conditions, the sums of the segments AB + BC and A′B′ + B′C′ can be the same for various values of the aspect angle γ. This equality of projections of the resulting velocity is kept constant with a given accuracy within the limits of some interval of values of the aspect angle γ covering the directional diagram width. By virtue of this fact, the total power spectral density of the target return signal ﬂuctuations contracts to a discrete line.

Copyright 2005 by CRC Press

1598_book.fm Page 287 Monday, October 11, 2004 1:57 PM

Fluctuations Caused by Moving Scatterers

287

gh′ Vr gh′

A′ B′ C′ gh Vr A C

gh

B

FIGURE 7.2 The power spectral density narrowing with the layered wind.

7.1.3

The Radar Antenna Is Directed along the Line of the Moving Radar

In this case, the conditions β0 = 0° and γ0 = 0° are satisﬁed. Assume that ∆Vw 0 = 0. Then, we can write

Ω′0 = 4πλ −1Vr ;

(7.21)

∆F = 2 λ −1 ∆ h | g|ρ cos β w .

(7.22)

Reference to Equation (7.21) and Equation (7.22) shows that if the condition βw ≠ 90° is satisﬁed, the effective bandwidth of the power spectral density of the target return signal ﬂuctuations is not equal to zero. Unlike in Equation (3.81), the formula in Equation (7.22) is true in the case when the radar antenna is directed exactly along the line of the moving radar. This phenomenon can be explained by the fact that with a given position of the radar antenna, the effective power spectral density bandwidth of the ﬂuctuations caused by stimulus of the layered wind is much more than that caused only by the moving radar. Because of this, we can neglect the target return signal ﬂuctuations caused by the moving radar. m km; Let us consider an example. Let ∆h = 2°, λ = 3 cm, ρ = 10 km, | g|= 2 sec this value is universally adopted for the middle latitudes. Then, we obtain that ∆F ≈ 30 Hz. This value is a whole order of magnitude greater than the effective bandwidth of the power spectral density of the ﬂuctuations caused Copyright 2005 by CRC Press

1598_book.fm Page 288 Monday, October 11, 2004 1:57 PM

288

Signal and Image Processing in Navigational Systems

only by the moving radar if the radar antenna is directed on the line of the moving radar [see Equation (3.141)]. However, we must bear in mind that these values depend on various parameters. Because of this, we must check this conclusion for every individual case in practice. If the condition ∆Vw ≠ 0 0 is satisﬁed, then, as follows from Equation (7.16), the effective power spectral density bandwidth of the ﬂuctuations of the target return signal is not equal to zero even if the condition βw = 90° is satisﬁed. 7.1.4

The Stationary Radar

If the condition

Va = Vr + Vw0 ≡ 0

(7.23)

is true, the radar is stationary. Reference to Equation (7.23) shows that Vr = –Vw′ and βw =0°. Let us recall that the angle βw is reckoned from the direction 0 of the vector Vr . Under these conditions, we obtain that

Ω′0 = 4πλ −1[Vw0 +| g|ρ sin γ 0 ] cos β 0 cos γ 0 ;

(7.24)

∆F = 2 λ −1

× ∆ (h2 ) (Vw0 +| g|ρ sin γ 0 )2 sin 2 β 0 + ∆ (v2 ) (Vw0 sin γ 0 −| g|ρ cos 2 γ 0 )2 cos 2 β 0 (7.25) Formulae in Equation (7.24) and Equation (7.25) are the main ones that consider the deterministic wind in navigational systems. In particular, taking into consideration Equation (7.24) and Equation (7.25), we can show that if the directional diagram axis is deployed along the direction of the wind, i.e., the condition β0 = 0° is satisﬁed, the ﬁrst term in Equation (7.25) is equal to zero. Then, with low values of the aspect angle γ0 , the layered wind plays the main role to form the power spectral density of the target return signal ﬂuctuations. If the directional diagram axis is perpendicular to the direction of the wind, the second term in Equation (7.25) is equal to zero. Then, with low values of the aspect angle γ0 , the velocity Vw of the wind plays the main 0 role. In some cases, Equation (7.25) can be used to measure the gradient g of the wind velocity. Under the conditions β0 = 0° or β0 = ±0.5π, we can choose such values of the aspect angle γ0 and the parameter ρ that satisfy the condition ∆F = 0 — the target return signal must be gated by the radar range. Then, knowing the values of the aspect angle γ0 , the parameter ρ, and the velocity Vw0 , we can easily determine the magnitude of the gradient g of the wind velocity.

Copyright 2005 by CRC Press

1598_book.fm Page 289 Monday, October 11, 2004 1:57 PM

Fluctuations Caused by Moving Scatterers

7.2

289

Scatterers Moving Chaotically (Displacement and Rotation)

Let us consider the target return signal ﬂuctuations with the chaotic motion of scatterers, taking into account both changes in the phase of elementary signals caused by the radial displacements of scatterers and changes in the amplitudes of elementary signals caused by the rotation of scatterers. We neglect changes in phases of the elementary signals caused by the rotation of scatterers and changes in the amplitudes of elementary signals caused by the radial displacements of scatterers.10–13 We assume that the radial displacements ∆ρ and the angle shifts ∆ξ and ∆ζ of scatterers are random variables and that the amplitude S of the target return signal is also a random variable. We assume that the joint probability distribution density f(∆ρ, ∆ξ, ∆ζ, S) is known. In the general case, these values can be dependent. Let us then assume that the joint probability distribution density f(∆ρ, ∆ξ, ∆ζ, S) is the same for all scatterers and is independent of the coordinates ρ, ϕ, and ψ. Because the number of scatterers is very high, the magnitude of f(∆ρ, ∆ξ, ∆ζ, S)d (∆ρ)d (∆ξ)d (∆ζ)dS characterizes the number of scatterers, in which the radial displacements, angle shifts, and amplitudes of the target return signal are within the limits of the corresponding intervals [∆ρ, (∆ρ + d(∆ρ)], [∆ξ, ∆ξ + d(∆ξ)], [∆ζ, ∆ζ + d(∆ζ)], and [S, S + dS]. At ﬁrst, we consider the totality of scatterers, in which the radial displacements, angle shifts, and amplitudes of the target return signal are within the limits of the intervals just mentioned. Then, we can use Equation (2.74), in which it is necessary to assume τ = nTp because we consider only the slow ﬂuctuations, ∆ϕ = 0°, ∆ψ = 0°, and ∆ρ = 0 for the argument of the pulse function P, and the radial displacements ∆ρ in exponent are independent of the angles ϕ and ψ. Then, without determining exactly the power of the target return signal, the correlation function of the ﬂuctuations for the considered totality can be written in the following form:

R′(∆ρ, ∆ξ , ∆ζ) ≅ S2 R′(∆ρ, ∆ξ , ∆ζ) × f (∆ρ, ∆ξ , ∆ζ, S) d(∆ρ) d(∆ξ) d(∆ζ) dS,

(7.26)

where

R′(∆ρ, ∆ξ , ∆ζ) = N ⋅

∫ ∫ q(ξ, ζ) ⋅ q(ξ + ∆ξ, ζ + ∆ζ) ⋅ e

− 2 jω 0

∆ρ c

sin ζ dξ dζ ; (7.27)

S2 is the value that is proportional to the power of the target return signal;

Copyright 2005 by CRC Press

1598_book.fm Page 290 Monday, October 11, 2004 1:57 PM

290

Signal and Image Processing in Navigational Systems

f (∆ρ, ∆ξ , ∆ζ, S)d(∆ρ)d(∆ξ)d(∆ζ)dS

(7.28)

is the averaged number of scatterers for the given totality. The resulting correlation function of the ﬂuctuations from all scatterers for arbitrary values of ∆ρ, ∆ξ, ∆ζ, and S in accordance with Equation (2.17) can be written in the following form: ∞ ∞ ∞ ∞

R∆enρ, ∆ξ , ∆ζ (∆ρ, ∆ξ , ∆ζ, S) ≅

∫ ∫ ∫ ∫ S R′(∆ρ, ∆ξ, ∆ζ) , 2

−∞ −∞ −∞ 0

(7.29)

× f (∆ρ, ∆ξ , ∆ζ, S) d(∆ρ) d(∆ξ) d(∆ζ) dS where integration with respect to the variables ∆ξ and ∆ζ is carried out within the limits of the interval [–∞, ∞] because there are no limitations for these variables. Based on Equation (7.29), the normalized correlation function of the ﬂuctuations has the following form:

R∆enρ, ∆ξ , ∆ζ (∆ρ, ∆ξ , ∆ζ, S) = ∞ ∞ ∞ ∞

∫ ∫ ∫ ∫ S R′(∆ρ, ∆ξ, ∆ζ) ⋅ f (∆ρ, ∆ξ, ∆ζ, S) d(∆ρ) d(∆ξ) d(∆ζ) dS , 2

−∞ −∞ −∞ 0

∞

∫ S f (S) dS 2

S

0

(7.30) where fs(S) is the probability distribution density of the amplitude of the target return signal. Hereafter, we assume that the angle shifts ∆ξ and ∆ζ are independent of the radial displacements ∆ρ and the amplitude S of the target return signal. Because of this, we can write

f (∆ρ, ∆ξ , ∆ζ, S) = f ∆ρ,S (∆ρ, S) ⋅ f ∆ξ , ∆ζ (∆ξ , ∆ζ) .

(7.31)

Reference to Equation (7.30) shows that

R∆enρ, ∆ξ , ∆ζ (∆ρ, ∆ξ , ∆ζ, S) = R∆ρ (∆ρ) ⋅ R∆ξ , ∆ζ (∆ξ , ∆ζ) , where

Copyright 2005 by CRC Press

(7.32)

1598_book.fm Page 291 Monday, October 11, 2004 1:57 PM

Fluctuations Caused by Moving Scatterers

291

∞ ∞

∫ ∫S f 2

R∆ρ (∆ρ) =

∆ρ,S

(∆ρ, S) ⋅ e

4 jπ

∆ρ λ

d(∆ρ) dS

−∞ 0

;

∞

(7.33)

∫ S f (S) dS 2

S

0

R∆ξ , ∆ζ (∆ξ , ∆ζ) = π π ∞ ∞

∫ ∫ ∫ ∫ q(ξ, ζ) ⋅ q(ξ + ∆ξ, ζ + ∆ζ) ⋅ f

∆ξ , ∆ζ

(∆ξ , ∆ζ)sin ζ d(∆ξ) d(∆ζ) dξ dζ

0 −π − ∞ − ∞

.

π π

∫ ∫ q (ξ, ζ)sin ζ dξ dζ 2

0 −π

(7.34) Thus, under the condition that the chaotic radial displacements and rotations of scatterers are the independent random variables, the normalized correlation function of the ﬂuctuations caused by simultaneous chaotic radial displacements and rotations of scatterers is deﬁned by the product of the normalized correlation function of the ﬂuctuations caused only by the radial displacements of scatterers and that caused only by the rotation of scatterers. Let us consider particular cases in the use of Equation (7.33) and Equation (7.34).

7.2.1

Amplitudes of Elementary Signals Are Independent of the Displacements of Scatterers

If the amplitude S of the target return signal — the target return signal is the sum of elementary signals, as before — and the radial displacements ∆ρ of scatterers are independent random variables, we can write

f ∆ρ,S (∆ρ, S) = f ∆ρ (∆ρ) ⋅ fS (S) .

(7.35)

Then, instead of Equation (7.25), the normalized correlation function of the target return signal ﬂuctuations caused only by the radial displacements of scatterers can be written in the following form: ∞

R∆ρ (∆ρ) =

∫

−∞

Copyright 2005 by CRC Press

f ∆ρ (∆ρ) ⋅ e

4 jπ

∆ρ λ

d(∆ρ) .

(7.36)

1598_book.fm Page 292 Monday, October 11, 2004 1:57 PM

292

Signal and Image Processing in Navigational Systems

This formula in Equation (7.36) coincides with the results discussed in Vanshtein and Zubakov.14 Therefore, we will consider it brieﬂy here. Let us assume that the random radial displacements ∆ρ of scatterers obey the Gaussian law, i.e., ∆ρ2

− 1 f ∆ρ (∆ρ) = ⋅ e 2σ , 2π σ ∆ρ 2 ∆ρ

(7.37)

2 is the variance of the radial displacements ∆ρ of scatterers. Under where σ∆ρ this assumption, the normalized correlation function of the ﬂuctuations caused only by the radial displacements ∆ρ of scatterers has the following form:

R∆ρ (∆ρ) = e

− 8π2

σ 2∆ρ λ2

.

(7.38)

To deﬁne the normalized correlation function R∆ρ(∆ρ) of the ﬂuctuations caused only by the radial displacements ∆ρ of scatterers, which is given by Equation (7.12), as a function of the shift τ in time, it is necessary to know 2 as a function of τ. the variance σ∆ρ It would appear reasonable that changes in the velocities of scatterers are low in value and can be considered approximately as constant values within short time intervals. Because of this, with low values of τ, we have r ∆ρ = Vscat ⋅τ

and

σ 2∆ρ = σ V2 scr at ⋅ τ 2 ,

(7.39)

r where Vscat is the radial component of the velocity of scatterers; σ V2 r is the scat

r variance of the ﬂuctuations of the radial component Vscat of the velocity of scatterers. Substituting Equation (7.39) in Equation (7.38), we can easily deﬁne the conditions in which we conclude that the velocities of scatterers are constant in the determination of the normalized correlation function of the ﬂuctuations R∆ρ(∆ρ) caused only by the radial displacements ∆ρ of scatterers, i.e., we can consider this stochastic process as a quasi-stationary process. If the condition in Equation (7.39) is true even for values of τ satisfying the condition

2 8 π 2 λ −2 σ V2 scat r τ > 4

or

σ Vsrcat ⋅ τ = σ ∆ρ > 0.25λ ,

(7.40)

in which R∆ρ(∆ρ), given by Equation (7.38) tends to approach zero, the velocities of scatterers can be considered constant if their variation is low in value

Copyright 2005 by CRC Press

1598_book.fm Page 293 Monday, October 11, 2004 1:57 PM

Fluctuations Caused by Moving Scatterers

293

within the limits of the time interval in which the root mean square deviation σ∆ρ of the radial displacements of scatterers becomes greater than the ratio 0.25λ. It seems likely that this condition is true for scatterers moving under the stimulus of the wind if the value λ is very low. In the case of constant or approximately constant velocities of the moving scatterer, the power spectral density of the ﬂuctuations caused only by the radial displacements (∆ρ) of scatterers takes the Gaussian shape after multiplying Equation (7.38) with the exponent ejω0τ and can be written in the following form:

S(ω) ≅ e

−π

( ω − ω 0 )2 ( 2 π ∆F∆ρ )2

.

(7.41)

The effective bandwidth of the power spectral density S(ω) of the ﬂuctuations caused only by the radial displacements (∆ρ) of scatterers, which is given by Equation (7.41), can be determined in Hz by −1 ∆F∆ρ = 8πσ Vscat ≈ 5σ V r λ −1 . r λ scat

(7.42)

The effective bandwidth ∆F∆ρ of the power spectral density S(ω) depends on the variance of the radial velocities of scatterers, which is the function of the characteristics of the wind and the aerodynamics of scatterers, and on the length of the wave λ.6,15,16 Under the same conditions of observation, ∆F∆ρ is higher and the wavelength λ is shorter, unlike the effective bandwidth of the power spectral density of the Doppler ﬂuctuations of the target return signal, which is independent of the wavelength λ with the deﬂected directional diagram.

7.2.2

The Velocity of Moving Scatterers Is Random but Constant

Let us assume that scatterers move with approximately constant (in the sense r elaborated in the previous section) but random velocities so that ∆ρ = Vscat · τ. We do not make any simple assumptions regarding the probability distribution density of the radial displacements or velocities of moving scatterers and the amplitude of elementary signals. Then, based on Equation (7.33), the normalized correlation function of the ﬂuctuations can be determined in the following form: ∞ ∞

R(τ) =

∫∫

r S2 fVscat , S) ⋅ e (Vscat r ,S

4 jπ

r Vscat ⋅τ λ

−∞ 0

.

∞

∫ S f (S) dS 2

S

0

Copyright 2005 by CRC Press

r dVscat dS

(7.43)

1598_book.fm Page 294 Monday, October 11, 2004 1:57 PM

294

Signal and Image Processing in Navigational Systems

Multiplying the normalized correlation function R(τ) of the ﬂuctuations by the exponent ejω0τ and using the Fourier transform, we can write the power spectral density of the ﬂuctuations in the following form: ∞

λ S(ω) ≅ ⋅ S2 fVscat [ λ (ω − ω 0 ), S] dS . r ,S 4 π 4π

∫

(7.44)

0

In particular, if the amplitude S of the target return signal and the radial r velocity Vscat of the radial displacements ∆ρ of scatterers are mutually independent, the normalized correlation function R(τ) and the power spectral density S(ω) of the ﬂuctuations can be determined in the following form: ∞

R(τ) =

∫f

r Vscat

(V ) ⋅ e r scat

4 jπ

r Vscat τ λ

r , dVscat

(7.45)

−∞

S(ω) ≅

λ ⋅ f r [ λ (ω − ω 0 ) ] 4π Vscat 4 π

(7.46)

instead of Equation (7.43) and Equation (7.44). The formulae in Equation (7.44) and Equation (7.46) have a simple physical meaning. Reference to Equation (7.46) shows that if the scatterers move with random but constant velocities and the amplitude S of the target return signal is independent of the velocity of moving scatterers, the power spectral density of the ﬂuctuations is deﬁned by the probability distribution density of radial velocities of moving scatterers if the condition r = 0.25π –1 λ(ω – ω0) Vscat

(7.47)

is satisﬁed in Equation (7.46). It is, naturally, because r ω − ω 0 = 4 π Vscat λ −1

(7.48)

r is the Doppler shift in frequency with the velocity Vscat of moving scatterers. r If the velocities Vscat of moving scatterers and the amplitude S of the target return signal are functionally related, the power spectral density of the ﬂuctuations is deﬁned by the joint probability distribution density of the r power of the target return signal and the radial velocities Vscat of moving scatterers. The formulae in Equation (7.44) and Equation (7.46) can be used successfully to determine the power spectral density of the ﬂuctuations if these r are caused by moving scatterers and the radial velocities Vscat of moving

Copyright 2005 by CRC Press

1598_book.fm Page 295 Monday, October 11, 2004 1:57 PM

Fluctuations Caused by Moving Scatterers

295

scatterers are constant or vary slowly. For example, the power spectral density of the ﬂuctuations, which is given by Equation (7.41), follows from r Equation (7.46) if we use the probability distribution density of Vscat based on the probability distribution density of the radial displacements ∆ρ of scatterers [see Equation (7.37)] using Equation (7.39). We will discuss further that the formulae in Equation (7.44) and Equation (7.46) can be used in the r cases where the radial velocities Vscat of moving scatterers are not random variables but are deﬁnite functions of space coordinates.

7.2.3

The Amplitude of the Target Return Signal Is Functionally Related to Radial Displacements of Scatterers

This case, in a certain sense, is the opposite of the case discussed in Section 7.2.1. Let us assume that function S = U(∆ρ) exists between the radial displacements ∆ρ of scatterers and the amplitude S of the target return signal. The joint probability distribution density takes the following form:

f ∆ρ,S (∆ρ, S) = f ∆ρ (∆ρ) ⋅ fS (S|∆ρ),

(7.49)

where fs(S|∆ρ) is the conditional probability distribution density of the amplitude S of the target return signal under the condition that the radial displacements ∆ρ of scatterers takes a given value. Owing to the strong functional relationship between S and ∆ρ, we can write

fS (S|∆ρ) = δ[S − U (∆ρ)] .

(7.50)

Then, based on Equation (7.33) the normalized correlation function R∆ρ(∆ρ) of the ﬂuctuations caused only by the radial displacements ∆ρ of scatterers can be determined by the following form: ∞

R∆ρ (∆ρ) =

∫

U 2 (∆ρ) f ∆ρ (∆ρ) ⋅ e

4 jπ

∆ρ λ

d(∆ρ)

−∞

.

∞

∫ U (∆ρ) f 2

∆ρ

(7.51)

(∆ρ) d(∆ρ)

−∞

The formula in Equation (7.51) can be applied in the case of the deterministic motion of scatterers when the power of the searching signal and the radial displacements ∆ρ are hardly related by the function. This function can appear, for example, when scatterers making radial displacements ∆ρ are localized within the deﬁnite volume of space and illuminated by the deﬁnite part of the directional diagram. It occurs if scatterers with the various

Copyright 2005 by CRC Press

1598_book.fm Page 296 Monday, October 11, 2004 1:57 PM

296

Signal and Image Processing in Navigational Systems

r effective scattering areas move with the different radial velocities Vscat . For example, large raindrops have a more effective scattering area and fall faster in comparison with small rain drops; passive scatterers of various dimensions and forms move with different velocities.17 The shape of the function U(∆ρ) depends on the shape of the directional diagram and on the law of moving scatterers. Using Equation (7.51), we can solve the problem considered here before dealing with problems concerning the moving radar and scatterers under the stimulus of the regular wind. The formula in Equation (3.8) was used to solve the problem of the deterministic motion of scatterers, and it has the same physical content as the formula in Equation (7.51), and both can be transformed by each other. These formulae are consequences of the general formula given by Equation (2.74). For this reason, we do not further investigate the formula in Equation (7.51) in this section.

7.2.4

Chaotic Rotation of Scatterers

Let us consider the normalized correlation function R∆ξ,∆ζ(∆ξ, ∆ζ) of the ﬂuctuations caused only by the rotation of scatterers under the assumption that the half-wave dipoles are scatterers. As was discussed in Section 5.4, the following deﬁnition q(ξ, ζ) = cos2 ξsin2 ζ is true for the component of the target return signal matched by polarization with the searching signal. Then, under the condition that the variables ∆ξ and ∆ζ are independent of the variables ξ and ζ, based on Equation (7.34), R∆ξ,∆ζ(∆ξ, ∆ζ) takes the following form: ∞

[ ∫f

1 R∆ξ , ∆ζ (∆ξ , ∆ζ) = R∆ξ (∆ξ)R∆ζ (∆ζ) = 2 + 9 ∞

[ ∫f

× 2+

∆ζ

∆ξ

]

(∆ξ)cos 2 ∆ξ d(∆ξ)

−∞

.

]

(∆ζ)cos 2 ∆ζ d(∆ζ)

−∞

(7.52) If the variables ∆ξ and ∆ζ (rotation of scatterers) obey the Gaussian law with zero mean and the variances σ 2∆ξ and σ 2∆ζ , respectively, the normalized correlation function R∆ξ,∆ζ(∆ξ, ∆ζ) is determined by

R∆ξ , ∆ζ (∆ξ , ∆ζ) ≅ 0.11 ⋅ [2 + e − 2σ

2 ∆ξ

][2 + e − 2σ ] . 2 ∆ζ

(7.53)

If the random angular velocities Ωξ and Ωζ of scatterer rotations are slowly varied, we can deﬁne

∆ξ = Ωξ ⋅ τ and ∆ζ = Ωζ ⋅ τ ;

Copyright 2005 by CRC Press

(7.54)

1598_book.fm Page 297 Monday, October 11, 2004 1:57 PM

Fluctuations Caused by Moving Scatterers

297

σ 2∆ξ = σ Ω2 ξ ⋅ τ 2 and σ 2∆ζ = σ Ω2 ζ ⋅ τ 2 ,

(7.55)

where σ Ω2 ξ is the variance of the angular velocity Ωξ of the rotation ∆ξ of scatterers; σ Ω2 ζ is the variance of the angular velocity Ωζ of the rotation ∆ζ of scatterers. As in the previous section, we deﬁne the conditions in which the angular velocities Ωξ and Ωζ can be considered the constant values. In this case, we obtain

σ 2∆ξ = σ Ω2 ξ ⋅ τ 2 > 2 and σ 2∆ζ = σ Ω2 ζ ⋅ τ 2 > 2

(7.56)

instead of Equation (7.40). This means that the angular velocities Ωξ and Ωζ of scatterers can be considered constant if we can neglect their variation within the limits of the time interval, in which the root mean square deviations σ∆ξ and σ∆ζ of the rotations ∆ξ and ∆ζ of scatterers become more than 90°, i.e., 2 radian. Substituting Equation (7.55) and Equation (7.56) in Equation (7.53), and multiplying after substitution on the exponent ejω0τ, and using the Fourier transform, the power spectral density of the ﬂuctuations caused only by the chaotic rotations ∆ξ and ∆ζ of scatterers can be written in the following form: −π

S(ω) ≅ 4δ(ω − ω 0 ) +

e

( ω − ω 0 )2 ∆Ω2∆ξ + ∆Ω2∆ζ

∆Ω2∆ξ + ∆Ω2∆ζ

−π

+ 2⋅

e

( ω − ω 0 )2 ∆Ω2∆ξ

∆Ω ∆ξ

−π

+ 2⋅

e

( ω − ω 0 )2 ∆Ω2∆ζ

∆Ω ∆ζ

, (7.57)

where

∆Ω ∆ξ = 2 π ∆F∆ξ = 8π σ Ωξ

and

∆Ω ∆ζ = 2 π ∆F∆ζ = 8π σ Ωζ . (7.58)

Thus, the power spectral density S(ω) of the ﬂuctuations caused only by the chaotic rotations ∆ξ and ∆ζ of scatterers consists of the one discrete line at the frequency ω0 of the searching signal and the continuous power spectral density in the form of three Gaussian power spectral densities (see Figure 7.3). The effective bandwidth of each Gaussian power spectral density is equal to ∆F∆ξ , ∆F∆ζ , and ∆F∆2ξ + ∆F∆2ζ , respectively. Reference to Equation (7.53) shows that 4/9 of the total power of the target return signal is concentrated in the discrete component of the power spectral density S(ω) of the ﬂuctuations caused only by the chaotic rotations ∆ξ and ∆ζ of scatterers at the frequency ω0; 5/9 is concentrated in the continuous Copyright 2005 by CRC Press

1598_book.fm Page 298 Monday, October 11, 2004 1:57 PM

298

Signal and Image Processing in Navigational Systems S (ω)

1 2

3

4 ω ω0 FIGURE 7.3 The power spectral density. Chaotic rotation of scatterers: (1) δ(ω – ω0); (2) S∆ξ(ω); (3) S∆ζ(ω); (4) S∆ξ,∆ζ(ω).

part. If the scatterers are stationary, the total power of the target return signal is concentrated near the discrete line of S(ω) caused by the radial displacements ∆ρ of scatterers and ∆ξ and ∆ζ at the frequency ω0. The chaotic rotations ∆ξ and ∆ζ lead us to the generation of the continuous part of S(ω) caused by ∆ρ and ∆ξ and ∆ζ at the frequency ω0, which possesses 5/9 of the total power of the target return signal independent of the shape of the probability distribution density. Unlike this, the chaotic motion (the radial displacements ∆ρ and chaotic rotations ∆ξ and ∆ζ) of scatterers leads to the spread power spectral density of the ﬂuctuations in which all discrete components are absent and the total power of the target return signal is concentrated in the continuous power spectral density of ﬂuctuations.

7.2.5

Simultaneous Chaotic Displacements and Rotations of Scatterers

Comparing Equation (7.42) and Equation (7.58), we can see that if the scatterers moving with approximately the same linear and angular velocities make one rotation on average within the limits of the segment with length equal to the length of the wave λ, i.e., when the condition

σ r V

scat

λ

≈

σΩ 2π

is satis-

ﬁed, then the chaotic displacements ∆ρ and rotations ∆ξ and ∆ζ of scatterers form the power spectral densities of the ﬂuctuations with the same effective bandwidths. Because with the simultaneous chaotic displacements ∆ρ and rotations ∆ξ and ∆ζ of scatterers the normalized correlation function R∆ρ(∆ρ) of the ﬂuctuations caused only by the radial displacements ∆ρ of scatterers, which is given by Equation (7.33) and the normalized correlation function R∆ξ,∆ζ(∆ξ,

Copyright 2005 by CRC Press

1598_book.fm Page 299 Monday, October 11, 2004 1:57 PM

Fluctuations Caused by Moving Scatterers

299

∆ζ) of the ﬂuctuations caused only by the rotations ∆ξ and ∆ζ of scatterers, which is given by Equation (7.34), are multiplied by each other, then, with slowly varying linear and angular velocities, the resulting normalized correlation function R∆ρ,∆ξ,∆ζ(∆ρ, ∆ξ, ∆ζ) of the ﬂuctuations caused by simultaneous chaotic radial displacements ∆ρ and rotations ∆ξ and ∆ζ is deﬁned by the product of R∆ρ(∆ρ), which is given by Equation (7.38), and R∆ξ,∆ζ(∆ξ, ∆ζ), which is given by Equation (7.53). Thus, Equation (7.39), Equation (7.54), and Equation (7.55) must be satisﬁed in Equation (7.38) and Equation (7.53), respectively. The resulting normalized correlation function R∆ρ,∆ξ,∆ζ(∆ρ, ∆ξ, ∆ζ) can be determined by the following form: σ2

R∆ρ, ∆ξ , ∆ζ (∆ρ, ∆ξ , ∆ζ) ≅ 0.11 ⋅ [2 + e −

2σ 2∆ξ

⋅τ

2

][2 + e −

2σ 2∆ζ

⋅τ

2

] ⋅e

− 8π

r Vscat 2

λ

τ2

. (7.59)

The resulting power spectral density of the ﬂuctuations caused by simultaneous chaotic radial displacements ∆ρ and rotations ∆ξ and ∆ζ of scatterers, which is shifted by the frequency ω0 , has the following form:

S(ω) = S1 (ω ) + S2 (ω ) + S3 (ω ) + S4 (ω ) ,

(7.60)

where −π 4 S1 (ω) ≅ ⋅e ∆Ω ∆ρ

S2 (ω) ≅

S3 (ω) ≅

S4 (ω) ≅

2 ∆Ω2∆ρ + ∆Ω2∆ξ 2 ∆Ω2∆ρ + ∆Ω2∆ζ

( ω − ω 0 )2 ∆Ω2∆ρ

⋅e

⋅e

1 ∆Ω2∆ρ + ∆Ω2∆ξ + ∆Ω2∆ζ

−π

−π

⋅e

∆Ω ∆ρ = 2 π ∆F∆ρ

;

(7.61)

( ω − ω 0 )2 ∆Ω2∆ρ + ∆Ω2∆ξ

;

(7.62)

;

(7.63)

( ω − ω 0 )2 ∆Ω2∆ρ + ∆Ω2∆ζ

−π

( ω − ω 0 )2 ∆Ω2∆ρ + ∆Ω2∆ξ + ∆Ω2∆ζ

;

(7.64)

(7.65)

is determined by Equation (7.42); ∆Ω∆ξ and ∆Ω∆ζ are given by Equation (7.58).

Copyright 2005 by CRC Press

1598_book.fm Page 300 Monday, October 11, 2004 1:57 PM

300

Signal and Image Processing in Navigational Systems

The resulting power spectral density S(ω), which is given by Equation (7.60), consists of four overlapping continuous power spectral densities containing 4/9, 2/9, 2/9, and 1/9 of the total target return signal power, respectively, and takes the Gaussian shape. The effective bandwidths of each component of S(ω) are determined in Hz by ∆F∆ρ,

∆F∆2ρ + ∆F∆2ξ ,

∆F∆2ρ + ∆F∆2ζ , and

∆F∆2ρ + ∆F∆2ξ + ∆F∆2ζ , respectively (see Figure 7.4). Thus, we can conclude as follows. If the chaotic rotations ∆ξ and ∆ζ of scatterers are slow, i.e., their rotations on average are much less than 2π with the radial displacements ∆ρ on the wavelength λ, then the following condition ∆F∆ξ , ∆F∆ζ << ∆F∆ρ is true, and we can neglect the chaotic rotations ∆ξ and ∆ζ of scatterers. In this case, the power spectral density S(ω) of the target return signal ﬂuctuations is Gaussian and can be determined by Equation (7.41). If ∆ξ and ∆ζ are fast, i.e., their rotations on average are much more than 2π with the radial displacements ∆ρ on the wavelength λ, then the following condition ∆F∆ξ, ∆F∆ζ >> ∆F∆ρ is true. In this case, S(ω) is deﬁned by the sum of four Gaussian power spectral densities: the narrow power spectral density S1(ω) with the effective bandwidth equal to ∆F∆ρ, and three wide power spectral densities S1(ω), S2(ω), and S3(ω). In the limiting case, as ∆F∆ρ → 0 and S1(ω) → δ(ω – ω0), the resulting S(ω) of the target return signal ﬂuctuations caused by simultaneous chaotic radial displacements ∆ρ and rotations ∆ξ and ∆ζ of scatterers tends to approach Equation (7.57). S (ω)

1

2 3

4 ω ω0 FIGURE 7.4 The power spectral density. Simultaneous chaotic displacements and rotations of scatterers: (1) S1(ω); (2) S2(ω); (3) S3(ω); (4) S4(ω).

Copyright 2005 by CRC Press

1598_book.fm Page 301 Monday, October 11, 2004 1:57 PM

Fluctuations Caused by Moving Scatterers

7.3 7.3.1

301

Simultaneous Deterministic and Chaotic Motion of Scatterers Deterministic and Chaotic Displacements of Scatterers

In the study of the simultaneous deterministic and chaotic displacements ∆ρ of scatterers without rotation, we should follow from the fact that Equation (7.33) is true for both cases. If scatterers take part in both forms of displacements, we can write18–20

∆ρ = ∆ρd + ∆ρch ,

(7.66)

where ∆ρd and ∆ρch are the displacements of scatterers caused by the deterministic motion and the chaotic motion, respectively. In the general case, the amplitude S of the target return signal is functionally related to both displacements. The character of this function has a pronounced effect on the shape of the resulting normalized correlation function of the ﬂuctuations. Let us consider the simplest particular case. Let us assume that the amplitude S of the target return signal is independent of both ∆ρd and ∆ρch . This case can occur with the rectangle directional diagram or if the width of the directional diagram is much more than the angle dimensions of the cloud of scatterers. It can also occur if scatterers with deﬁnite displacements ∆ρ have equiprobable distribution in space. Reference to Equation (7.36), which is true with the independent amplitudes S of the target return signal and displacements ∆ρ of scatterers, shows that to determine the normalized correlation function of the ﬂuctuations, it is necessary to know the probability distribution density of the displacements ∆ρ of scatterers given by Equation (7.66). Because the deterministic displacements ∆ρd and the chaotic displacements ∆ρch of scatterers are mutually independent the probability distribution density of simultaneous deterministic and chaotic displacements ∆ρ of scatterers can be deﬁned by the convolution between the probability distribution density of ∆ρd and ∆ρch in the following form:

f ∆ρ (∆ρ) = f ∆ρd (∆ρd ) ∗ f ∆ρch (∆ρch ) .

(7.67)

Substituting Equation (7.67) in Equation (7.36), we can prove that with the independent values of S and ∆ρ, the normalized correlation function of the ﬂuctuations R∆ρ(∆ρ) caused simultaneously by the deterministic ∆ρd and chaotic ∆ρch displacements is deﬁned by the product of two normalized correlation functions of the target return signal: that of R∆ρd(∆ρd) caused only by ∆ρd and that of R∆ρch(∆ρch) caused only by ∆ρch. R∆ρ(∆ρ) can be written in the following form:

Copyright 2005 by CRC Press

1598_book.fm Page 302 Monday, October 11, 2004 1:57 PM

302

Signal and Image Processing in Navigational Systems

R∆ρ (∆ρ) = R∆ρd (∆ρd ) ⋅ R∆ρch (∆ρch ) .

(7.68)

The normalized correlation function of the ﬂuctuations R∆ρd(∆ρd) caused only by the deterministic displacements ∆ρd and the normalized correlation function R∆ρch(∆ρch) caused only by the chaotic displacements ∆ρch can be determined based on Equation (7.33) if we change the displacements ∆ρ in ∆ρd and ∆ρch, respectively. Consider the general case, when the amplitude S of the target return signal is functionally related to the deterministic ∆ρd and chaotic ∆ρch displacements. To determine the normalized correlation function R∆ρ(∆ρ) caused simultaneously by the deterministic ∆ρd and chaotic ∆ρch , which is given by Equation (7.33), it is necessary to ﬁrst deﬁne the joint probability distribution density of the variables S and ∆ρ, where the displacements ∆ρ are given by Equation (7.66). Naturally, we assume that the functional relation between the amplitude S and the displacements ∆ρd and ∆ρch is known. In other words, we assume that the probability distribution densities f∆ρd ,S(∆ρd, S) and f∆ρch ,S(∆ρch, S) are known. Then, we can write

f ∆ρ,S (∆ρ, S) = fS (S) ⋅ f ∆ρ (∆ρ|S) ,

(7.69)

where f∆ρ(∆ρ|S) is the conditional probability distribution density of the displacements ∆ρ of scatterers. As is well known, the conditional probability distribution density is subject to the same laws as the usual (unconditional) probability distribution density. Because of this, we can write

f ∆ρ (∆ρ|S) = f ∆ρ [(∆ρd + ∆ρch )|S] = f ∆ρd (∆ρd |S) ∗ f ∆ρch (∆ρch |S) ∞

=

∫

. (7.70)

f ∆ρd (∆ρd |S) ⋅ f ∆ρch [(∆ρ − ∆ρd )|S] d(∆ρd )

−∞

Substituting Equation (7.70) in Equation (7.33) and Equation (7.69), ∞

∞

R∆ρ (∆ρ) =

∫ 0

S2 fS (S)

∫

f ∆ρ (∆ρ|S) ⋅ e

4 jπ

−∞

∆ρ λ

d(∆ρ) dS .

∞

(7.71)

∫ S f (S) dS 2

S

0

Note that the inside integral in the numerator of the formula in Equation (7.71) coincides with Equation (7.36). The difference is only that the conditional probability distribution density f∆ρ(∆ρ|S) is subjected to the Fourier Copyright 2005 by CRC Press

1598_book.fm Page 303 Monday, October 11, 2004 1:57 PM

Fluctuations Caused by Moving Scatterers

303

transform, and not the unconditional probability distribution density. Let us consider that the conditional probability distribution density is deﬁned by the convolution given by Equation (7.70) between the two conditional probability distribution densities, as well as the unconditional probability distribution density given by Equation (7.67). Let us introduce a designation ∞

∫

R∆ρ | S (∆ρ|S) =

f ∆ρ (∆ρ|S) ⋅ e

4 jπ

∆ρ λ

d(∆ρ)

(7.72)

−∞

that is similar to Equation (7.36). We call the function R∆ρ|S(∆ρ|S) the conditional normalized correlation function of the ﬂuctuations caused simultaneously by the deterministic ∆ρd and chaotic ∆ρch displacements of scatterers. Then we can write

R∆ρ | S (∆ρ|S) = R∆ρd|S (∆ρd |S) ⋅ R∆ρch|S (∆ρch |S) .

(7.73)

Substituting Equation (7.73) in Equation (7.71), in the general case, we see that the normalized correlation function of the ﬂuctuations R∆ρ(∆ρ) caused simultaneously by the deterministic ∆ρd and chaotic ∆ρch displacements of scatterers can be determined by the following form: ∞

∫ S f (S) ⋅ R 2

S

R∆ρ (∆ρ) =

∆ρd|S

(∆ρd |S) ⋅ R∆ρch|S (∆ρch |S) dS

0

,

∞

(7.74)

∫ S f (S) dS 2

S

0

i.e., Equation (7.68) is not true. However, if we assume that ∆ρd or ∆ρch are independent of the amplitude S, Equation (7.68) becomes true. From the preceding discussion particularly, it follows that Equation (7.68) is true if S is functionally related, for example, with ∆ρd and is independent of ∆ρch. This circumstance allows us to deﬁne R∆ρ(∆ρ) caused simultaneously by ∆ρd and ∆ρch based on the results discussed in the previous sections. Consider an example where the scatterers make simultaneously the deterministic motion caused by the moving radar and the layered wind and chaotic motion with velocities, which are distributed by the Gaussian law and varied slowly. The resulting normalized correlation function of the ﬂuctuations is deﬁned by the product of the normalized correlation functions given by Equation (3.79) and Equation (7.36). Therefore, Equation (7.37) must be substituted in Equation (7.36). The resulting power spectral density of the ﬂuctuations caused simultaneously by the deterministic ∆ρd and chaotic ∆ρch displacements of scatterers is deﬁned by the convolution between the power spectral densities given by Copyright 2005 by CRC Press

1598_book.fm Page 304 Monday, October 11, 2004 1:57 PM

304

Signal and Image Processing in Navigational Systems

Equation (3.80) and Equation (7.41). Because both these power spectral densities obey the Gaussian law, the resulting power spectral density caused simultaneously by ∆ρd and ∆ρch takes the Gaussian shape with the effective bandwidth determined by

∆F = ∆F∆2ρd + ∆F∆2ρch ,

(7.75)

where ∆F∆ρd is the effective bandwidth of the ﬂuctuations caused only by ∆ρd which is given by Equation (7.16), and ∆F∆ρch caused only by ∆ρch. 7.3.2

Chaotic Rotation of Scatterers and Rotation of the Polarization Plane

First, let us prove that by using Equation (7.34) and Equation (7.52) we can deﬁne the normalized correlation function of the ﬂuctuations caused by the rotation of the radar antenna polarization plane. Because with rotation of the radar antenna polarization plane the angles ∆ξ are the same for all scatterers, the probability distribution density of the variable ∆ξ can be written in the following form:21,22

f ∆ξ (∆ξ) = δ(∆ξ − ∆ξ d ) ,

(7.76)

where ∆ξd is the angle of rotation of the radar antenna polarization plane during the time τ. Assuming that scatterers are the half-dipoles, we can substitute Equation (7.76) in Equation (7.52). After integration, we obtain the normalized correlation function of the ﬂuctuations caused by rotation of the radar antenna polarization plane in the following form:

Rq (∆ξ d ) ≅ 0.33 ⋅ (2 + cos 2 ∆ξ d ) .

(7.77)

As we can see, Equation (7.77) coincides with Equation (5.88). Now, consider the simultaneous chaotic rotation of scatterers and the rotation of the polarization plane of the radar antenna.23 For this case, we can write

∆ξ = ∆ξ d + ∆ξ ch ,

(7.78)

where ∆ξch and ∆ξd are the angles of rotation due to the simultaneous chaotic rotation of scatterers and rotation of the polarization plane of the radar antenna, respectively. Because the random variable ∆ξch is independent of the nonrandom variable ∆ξd , the probability distribution density f∆ξ(∆ξ) of

Copyright 2005 by CRC Press

1598_book.fm Page 305 Monday, October 11, 2004 1:57 PM

Fluctuations Caused by Moving Scatterers

305

the variable ∆ξ is deﬁned by the convolution between the probability distribution density of the chaotic rotation ∆ξch of scatterers and that of the rotation ∆ξd of the radar antenna polarization plane, and can be written in the following form:

f ∆ξ (∆ξ) = f ∆ξch (∆ξ ch ) ∗ f ∆ξd (∆ξ d ) = f ∆ξch (∆ξ − ∆ξ d ) .

(7.79)

In this particular case, when the random variable ∆ξch obeys the Gaussian law, we can write

f ∆ξ (∆ξ) =

1 ⋅e 2π σ ∆ξch

−

( ∆ξ − ∆ξ d )2 2 σ2

∆ξ ch

.

(7.80)

Substituting Equation (7.80) in Equation (7.34) and assuming that, as before, the random variables ∆ξ and ∆ζ are mutually independent, we can write the resulting normalized correlation function of the ﬂuctuations R∆ξ,∆ζ(∆ξ, ∆ζ) in the following form:

R∆ξ , ∆ζ (∆ξ , ∆ζ) ≅ 0.11 ⋅ [2 + cos 2 ∆ξ d ⋅ e

− 2 σ 2∆ξ

ch

][2 + e − 2 σ ] . 2 ∆ζ

(7.81)

Comparing Equation (7.81) with Equation (7.53) and Equation (7.77), we can see that the resulting R∆ξ,∆ζ(∆ξ, ∆ζ) given by Equation (7.81) is not deﬁned by the product of the normalized correlation functions R∆ξ,∆ζ(∆ξ, ∆ζ) and Rq(∆ξd) of the ﬂuctuations, which are given by Equation (7.53) and Equation (7.77), respectively. If scatterers rotate with slow varied velocities, i.e., the conditions

σ ∆ξch = σ Ωξ ⋅ τ

and

σ ∆ζ = σ Ωζ ⋅ τ

(7.82)

are satisﬁed, and the polarization plane of the radar antenna rotates uniformly with the angular velocity Ωd, i.e., the condition ∆ξd = Ωd · τ is satisﬁed, then the normalized correlation function R(τ) of the ﬂuctuations caused simultaneously by the chaotic rotation of scatterers and the rotation of the polarization plane of the radar antenna can be written in the following form:

R(τ) ≅ 0.11 ⋅ [2 + cos 2Ωd τ ⋅ e

− 2 σ Ω2 τ 2 ξ

][2 + e

− 2 σ Ω2 τ 2 ξ

].

(7.83)

Multiplying Equation (7.83) by the exponent ejω0τ and using the Fourier transform, the power spectral density of the ﬂuctuations caused simulta-

Copyright 2005 by CRC Press

1598_book.fm Page 306 Monday, October 11, 2004 1:57 PM

306

Signal and Image Processing in Navigational Systems

neously by the chaotic rotation of scatterers and the rotation of the polarization plane takes the following form: S(ω) ≅ 8δ(ω – ω0) + S1(ω) + S2(ω) + S3(ω),

(7.84)

where −π 4 S1 (ω) ≅ ⋅e ∆Ω ∆ζ

−π 2 S2 (ω) ≅ ⋅ e ∆Ω ∆ξ

[

S3 (ω) ≅

1 ∆Ω2∆ξ + ∆Ω2∆ζ

( ω − ω 0 )2 ∆Ω2∆ζ

( ω − ω 0 − 2 Ωd )2 ∆Ω2∆ξ

[

⋅ e

−π

+e

;

−π

( ω − ω 0 + 2 Ωd )2 ∆Ω2∆ξ

( ω − ω 0 − 2 Ωd )2 ∆Ω2∆ξ + ∆Ω2∆ζ

(7.85)

+e

−π

];

( ω − ω 0 + 2 Ωd )2 ∆Ω2∆ξ + ∆Ω2∆ζ

(7.86)

].

(7.87)

The power spectral density S(ω) of the ﬂuctuations caused simultaneously by the chaotic rotation of scatterers and the rotation of the radar antenna polarization plane, which is given by Equation (7.84), consists of the one discrete line and the continuous power spectral density S1(ω) at the frequency ω0 and two continuous power spectral densities S2(ω) and S3(ω) at the frequencies ω0 ± 2Ωd , respectively (see Figure 7.5): S2′ (ω) and S3′ (ω) at the frequency ω0 + 2Ωd , and S2″ (ω) and S3″ (ω) at the frequency ω0 – 2Ωd. The discrete component 8δ(ω – ω0) of S(ω) possesses 8/18 of the total power of the target return signal; 4/18 of the total power is concentrated in S1(ω) at the frequency ω0 with the effective bandwidth equal to ∆Ω∆ζ , and 2/18 in S2′ (ω) at the frequency ω0 – 2Ωd with the effective bandwidth equal to ∆Ω∆ξ ; 2/18 is concentrated in S2″ (ω) at the frequency ω0 – 2Ωd with the effective bandwidth equal to ∆Ω∆ξ; 1/18 in S3′(ω) at the frequency ω0 – 2Ωd with the effective bandwidth equal to

∆Ω2∆ξ + ∆Ω2∆ζ ; 1/18 in S3″ (ω) at the frequency

ω0 – 2Ωd with the effective bandwidth equal to

7.3.3

∆Ω2∆ξ + ∆Ω2∆ζ .

Chaotic Displacements of Scatterers and Rotation of the Polarization Plane

Because the normalized correlation function of the ﬂuctuations caused only by the rotation of the radar antenna polarization plane, as well as those caused only by the chaotic rotation of scatterers, can be determined by Equation (7.34),

Copyright 2005 by CRC Press

1598_book.fm Page 307 Monday, October 11, 2004 1:57 PM

Fluctuations Caused by Moving Scatterers

307

S (ω)

1 2

4

3

6

5 ω

ω0 − 2Ωd

ω0

ω0 − 2Ωd

FIGURE 7.5 The power spectral density. Simultaneous chaotic rotation of scatterers and rotation of the polarization plane: (1) δ(ω – ω0); (2) S1(ω); (3) S2′ (ω); (4) S2″ (ω); (5) S3′ (ω); (6) S3″ (ω).

and the normalized correlation function of the ﬂuctuations caused only by chaotic displacements of scatterers is given by Equation (7.33), in accordance with Equation (7.32), the resulting normalized correlation function is deﬁned by the product of the normalized correlation functions. Let us consider the simplest example. Let us assume that the chaotic displacements of scatterers are distributed according to the Gaussian law. Scatterers move with approximately constant velocities, and the polarization plane of the radar antenna rotates uniformly with the angular velocity Ωd.24,25 Using Equation (7.38), Equation (7.39), and Equation (7.77), and taking into consideration the condition ∆ξd = Ωd · τ, the resulting normalized correlation function R(τ) caused simultaneously by the chaotic displacements of scatterers and the rotation of the radar antenna polarization plane can be written in the following form: σ2

R(τ) ≅ 0.33 ⋅ (2 + cos 2Ωd τ) ⋅ e

− 8π

r Vscat 2

λ

τ2

.

(7.88)

The corresponding power spectral density S(ω) (see Figure 7.6) can be determined by the following form:

S(ω) ≅ 4S∆ρ (ω ) + S∆ρ (ω − 2Ωd ) + S∆ρ (ω + 2Ωd ) ,

(7.89)

where the power spectral density S∆ρ(ω) is given by Equation (7.41). In an analogous way, we can deﬁne the normalized correlation function and power spectral density of the ﬂuctuations with simultaneous deterministic

Copyright 2005 by CRC Press

1598_book.fm Page 308 Monday, October 11, 2004 1:57 PM

308

Signal and Image Processing in Navigational Systems

S (ω)

1

3

2

ω ω0 − 2Ωd

ω0

ω0 − 2Ωd

FIGURE 7.6 The power spectral density. Simultaneous chaotic displacements of scatterers and rotations of the polarization plane: (1) 4S∆ρ(ω); (2) S∆ρ(ω – 2Ωd); (3) S∆ρ(ω + 2Ωd).

displacements and chaotic rotations of scatterers, as in this case, and the formula in Equation (7.32) is also true.

7.4

Conclusions

In the course of investigation of the target return signal ﬂuctuations caused by moving scatterers under the stimulus of the wind, we can conveniently consider two motion components of the cloud of scatterers: deterministic motion with different velocities, which can be deﬁned by the nonstochastic function of coordinates and time (for example, variation of the velocity of the wind as a function of altitude [the layered wind] and the motion of the cloud of scatterers as a whole) and stochastic motion with the random velocity and velocity that is varied in time, in the general case. With the target return signal ﬂuctuations caused by deterministic displacements of scatterers, in particular by the simultaneous stimulus of the layered wind and moving radar, the power spectral density coincides in shape with the squared directional diagram by power in the plane passing through the plane ϕ (the horizon) under the angle κ′ given by Equation (7.13). However, this plane does not pass through the direction of the moving radar and the directional diagram axis, neither does it occur in the absence of the wind. The effective bandwidth depends on distance between the radar and the target. This phenomenon exists because, in accordance with the distance between the radar and the target, the

Copyright 2005 by CRC Press

1598_book.fm Page 309 Monday, October 11, 2004 1:57 PM

Fluctuations Caused by Moving Scatterers

309

directional diagram illuminates areas of the cloud of scatterers having various velocities as a function of altitude. With simultaneous chaotic radial displacements and rotations of scatterers, if these are the independent random variables, the normalized correlation function of the target return signal ﬂuctuations is deﬁned by the product of that caused only by the chaotic radial displacements of scatterers and that caused only by the rotation of scatterers. In the general case, because with simultaneous chaotic displacements and rotations of scatterers the normalized correlation function R∆ρ(∆ρ) of the ﬂuctuations caused only by the chaotic displacements of scatterers and R∆ξ,∆ζ(∆ξ, ∆ζ) of the ﬂuctuations caused only by the rotation of scatterers are multiplied with each other, then, with linear and angular velocities that vary very slowly, the resulting normalized correlation function R∆ρ,∆ξ,∆ζ(∆ρ, ∆ξ, ∆ζ) of the ﬂuctuations caused simultaneously by chaotic displacements and rotations of scatterers is deﬁned by the product of R∆ρ(∆ρ), which is given by Equation (7.38), and R∆ξ,∆ζ(∆ξ, ∆ζ),which is given by Equation (7.53). The power spectral density of the ﬂuctuations caused simultaneously by chaotic displacements and rotations of scatterers consists of four overlapping power spectral densities possessing 4/9, 2/9, 2/9, and 1/9 of the total power of the target return signal, respectively (see Figure 7.4). With the simultaneous deterministic and chaotic displacements (without rotations) of scatterers, in the general case, the resulting normalized correlation function of the ﬂuctuations takes a more complex form [see Equation (7.74)]. The resulting power spectral density is deﬁned by the convolution between the power spectral densities caused only by the deterministic displacements of scatterers and chaotic displacements (without rotation) of scatterers. With the simultaneous chaotic rotation of scatterers and rotation of the polarization plane of the radar antenna, the power spectral density consists of (see Figure 7.5) the one discrete line at the frequency ω0 containing 8/18 of the total power of the target return signal; the continuous power spectral density at the frequency ω0 containing 4/18 of the total power of the target return signal; two continuous power spectral densities at the frequencies ω0 ± 2Ωd, with each possessing 2/18 of the total power of the target return signal; and two continuous power spectral densities at the frequencies ω0 ± 2Ωd, with each possessing 1/18 of the total power of the target return signal. With chaotic displacements of scatterers and the rotation of the radar antenna polarization plane, the power spectral density of the target return signal ﬂuctuations consists of three continuous power spectral densities (see Figure 7.6): with the center at the frequency ω0 , possessing 2/3 of the total power of the target return signal; with the center at the frequency ω0 + 2Ωd , possessing 1/6 of the total power of the target return signal; and with the center at the frequency ω0 – 2Ωd, possessing 1/6 of the total power of the target return signal.

Copyright 2005 by CRC Press

1598_book.fm Page 310 Monday, October 11, 2004 1:57 PM

310

Signal and Image Processing in Navigational Systems

References 1. Melnik, A., Zubkovich, C., Stepanenko, V. et al., Radar Methods of Investigation of the Earth Surface, Soviet Radio, Moscow, 1980 (in Russian). 2. Lourtie, I. and Carter, G., Signal detectors for random ocean media, J. Acoust. Soc. Amer., Vol. 92, No. 3, 1992, pp. 1420–1427. 3. Takao, K., Fujita, H., and Nishi, T., An adaptive array under directional constraint, IEEE Trans., Vol. AP-21, No. 9, 1976, pp. 662–669. 4. Stepanenko, V., Radar in Meteorology, Hydrometizdat, Leningrad, 1973 (in Russian). 5. Hall, H., A new model for impulsive phenomena: application to atmosphericnoise communication channel, Technical Report 3412-8, Stanford University, Stanford, CA, August 1966. 6. Dulevich, V., Theoretical Foundations of Radar, Soviet Radio, Moscow, 1978 (in Russian). 7. Borkus, M., Energy spectrum of signals scattered by atmosphere aerosol particles caused by moving radar, Problems in Radio Electronics, Vol. OT, No. 1, 1977, pp. 43–50 (in Russian). 8. Ursin, B. and Bertenssen, K., Comparison of some inverse methods for wave propagation: layered media, in Proceedings of the IEEE, Vol. 74, 1986, pp. 389–400. 9. Giordano, A. and Haber, F., Modeling of atmospheric noise, Radio Sci., Vol. 7, No. 8, 1972, pp. 1101–1123. 10. Levanon, N., Radar Principles, John Wiley & Sons, New York, 1988. 11. Andersh, D., Lee, S., and Ling, H., XPATCH: a high frequency electromagnetic scattering prediction code and environment for complex three-dimensional objects, IEEE Antennas Propagat. Mag., Vol. 36, No. 1, 1994, pp. 65–69. 12. Yagle, A. and Frolik, J., On the feasibility of impulse reﬂection response data for the two-dimensional inverse scattering problem, IEEE Trans., Vol. AP-44, No. 8, 1996, pp. 1551–1564. 13. Rendas, M. and Monra, J., Ambiguity in radar and sonar, IEEE Trans., Vol. SP46, No. 2, 1998, pp. 294–305. 14. Vanshtein, L. and Zubakov, V., Picking out the Signals in Noise, Soviet Radio, Moscow, 1960 (in Russian). 15. Potter, L. and Moses, R., Attributed scattering centers for SAR ATR, IEEE Trans., Vol. IP-6, No. 1, 1997, pp. 79–91. 16. Foschini, G., Layered space-time architecture for wireless communication in a fading environment when using multi-element antennas, Bell Labs Tech. J., Vol. 1, No. 2, 1996, pp. 41–59. 17. Habibi-Ashraﬁ, F. and Mendel, J., Estimation of parameters in loss-less layered media systems, IEEE Trans., Vol. AC-27, No. 1, 1982, pp. 31–48. 18. Widrow, B., Mantley, P., Grifﬁths, L., and Goode, B., Adaptive antennas systems, in Proceedings of the IEEE, Vol. 55, No. 12, 1967, pp. 2143–2159. 19. Compton, R., Adaptive Antennas, Prentice Hall, Englewood Cliffs, NJ, 1988. 20. Pillai, S., Array Signal Processing, Springer-Verlag, New York, 1989. 21. Holm, W., Polarimetric fundamentals and techniques, in Principles of Modern Radar, J. Eaves and E. Reedy, Eds., Van Nostrand Reinhold, New York, 1987.

Copyright 2005 by CRC Press

1598_book.fm Page 311 Monday, October 11, 2004 1:57 PM

Fluctuations Caused by Moving Scatterers

311

22. Ulaby, F. and Elachi, C., Radar Polarimetry for Geoscience Applications, Artech House, Norwood, MA, 1990. 23. Drane, C., Positioning Systems: A Unified Approach, Springer-Verlag, New York, 1992. 24. Cloude, S., Polarimetric techniques in radar signal processing, Microwave J., Vol. 26, No. 7, 1983, pp. 119–127. 25. Guili, D., Polarization diversity in radar, in Proceedings of the IEEE, Vol. 74, No. 2, 1986, pp. 245–269.

Copyright 2005 by CRC Press

1598_book.fm Page 313 Monday, October 11, 2004 1:57 PM

8 Fluctuations under Scanning of the TwoDimensional (Surface) Target with the Continuous Frequency-Modulated Signal

8.1

General Statements

The continuous frequency-modulated searching signal, in accordance with Equation (2.54), can be written in the following, form:

W (t) = W0 (t) ⋅ e − j[ω 0t + Ψ (t )] .

(8.1)

The instantaneous frequency of the searching signal given by Equation (8.1) has the following form:

ω(t) = ω 0 +

dΨ(t) . dt

(8.2)

In accordance with the general formula in Equation (2.55), the correlation function of target return signal ﬂuctuations reﬂected by the two-dimensional (surface) target, which is modulated by frequency (or phase), has the following form:

R(t, τ) = p0 ⋅ e jω 0τ ×e

[ (

j Ψ t −

∫∫

g 2 (ϕ, ψ ) ⋅ e

2ρ c

∆ρ c

−

− 2 jω 0

∆ρ c

, )

(

+ 0.5 τ − Ψ t −

2ρ c

−

∆ρ c

− 0.5τ

(8.3)

)] dϕ dψ

where g˜ 2 (ϕ , ψ ) is the generalized radar antenna directional diagram given by Equation (4.14);

313 Copyright 2005 by CRC Press

1598_book.fm Page 314 Monday, October 11, 2004 1:57 PM

314

Signal and Image Processing in Navigational Systems

p0 =

PG02 λ2 . 64 π 3 h 2

(8.4)

Representation of the searching signal in the form given by Equation (8.1) and the correlation function of the ﬂuctuations in the form determined by Equation (8.3) is convenient if the function Ψ(t) is continuous for all values of t, as is the case with harmonic frequency modulation. If the function Ψ(t) or its derivative is discontinuous, which is characteristically the case in linear frequency modulation, it is worthwhile to express the frequency-modulated searching signal as a time periodic sequence of signals that is coherent from period to period,1 i.e., the value of ω0TM is multiplied by 2π: ∞

W (t) = W0 (t)

∑e

− j[ ω 0t + Ψ ( t − nTM )]

,

(8.5)

n= − ∞

where 0 < |t – nTM| < TM and TM is the period of modulation. In this case, we can write

R(t, τ) = p0 ⋅ e jω 0τ ∞

×

∞

∑ ∑e

∫∫

{ [

g 2 (ϕ, ψ ) ⋅ e

j Ψ t − kTM −

2ρ c

−

∆ρ c

− 2 jω 0

∆ρ c

]

[

+ 0.5( τ − nTM ) − Ψ t − kTM −

2ρ c

−

∆ρ c

]}

− 0.5( τ − nTM )

.

dϕ dψ

k = − ∞ n= − ∞

(8.6) The correlation function of the ﬂuctuations given in Equation (8.6) is analogous to the correlation function of pulsed target return signal sequence ﬂuctuations and is periodic, both with respect to t and with respect to τ. This correlation function deﬁnes the ﬂuctuations from period to period, within the limits of a period. If we introduce the function Π(t) into the integrand, the correlation function of the ﬂuctuations given by Equation (8.6) allows us to deﬁne the correlation function of pulsed target return signal ﬂuctuations under intrapulsed and interpulsed frequency modulation. In most cases of radar with frequency-modulated searching signals, the target return signal is multiplied by the searching signal, which is a heterodyne signal.2–5 Let us assume that the heterodyne signal with unit amplitude is shifted by the intermediate frequency ωim , which is much larger than the bandwidth of the searching signal. Henceforth, the correlation function and power spectral density of the target return signal ﬂuctuations are deﬁned under the condition ωim ≠ 0. For ωim = 0, we must use |Ω0 – nωM| instead of frequencies ωim + Ω0 – nωM in all formulae, as negative frequencies have no physical meaning. When we use the searching signal given by Equation (8.1), the heterodyne signal takes the following form:

Copyright 2005 by CRC Press

1598_book.fm Page 315 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

315

Wh (t) = e − j [(ω 0 − ωim )t + Ψ (t )] .

(8.7)

In this case, the transformed target return signal from the i-th scatterer located at a distance ρ from the radar can be written in the following form: at the time instant t – 0.5τ, we obtain

Wtri (t − 0.5τ) = W0i (t) ⋅ Wh (t − 0.5τ) ⋅ Wri* (t − 0.5τ − 2cρ +

∆ρ c

)

(8.8)

)]* ,

(8.9)

and at the time instant t + 0.5τ, we obtain

Wtr*i (t + 0.5τ) = W0i (t)[Wh (t + 0.5τ) ⋅ Wri* (t + 0.5τ − 2cρ −

∆ρ c

where Wtri(t) is the transformed target return signal from the i-th scatterer; Wri(t) is the target return signal from the i-th scatterer; and W*(t) is the signal conjugated with the signal W(t). Multiplying the target return signal from the i-th scatterer by the heterodyne signal Wh(t) does not change the random character of the target return signal phase. Because of this, the transformed target return signal Wtri(t) can be considered as a stochastic process, and we can use the general formula in Equation (2.55) to determine the correlation function of the ﬂuctuations

R(t, τ) = p0 ⋅ e jωimτ ×e

[

∫∫

g 2 (ϕ, ψ ) ⋅ e

(

j Ψ ( t − 0.5 τ ) − Ψ t − 0.5τ −

2ρ c

+

∆ρ c

)]

− 2 jω 0

⋅e

∆ρ c

[

(

− j Ψ ( t + 0.5τ ) − Ψ t + 0.5τ −

2ρ c

−

∆ρ c

)]

dϕ dψ. (8.10)

In an analogous way, if the signal W(t) is given by Equation (8.5), we can write the correlation function of the ﬂuctuations in the following form:

R(t, τ) = p0 ⋅ e jωimτ ∞

×

∞

∑∑

e

∫∫

g 2 (ϕ, ψ ) ⋅ e

{ [

− 2 jω 0

]

∆ρ c

[

j Ψ t − kTM − 0.5( τ − nTM ) − Ψ t − kTM − 0.5( τ − nTM ) −

2ρ c

+

∆ρ c

]}

.

(8.11)

k = − ∞ n= − ∞

×e

{ [

]

[

− j Ψ t − kTM + 0.5( τ − nTM ) − Ψ t − kTM + 0.5( τ − nTM ) −

2ρ c

−

∆ρ c

]}

dϕ dψ

The function Ψ(τ) of the heterodyne signal given by Equation (8.8) is taken within the limits of the same period k, as a shift in t by an integer does not change the target return signal due to the fact that the heterodyne signal is coherent. The formulae in Equation (8.10) and Equation (8.11), as well as

Copyright 2005 by CRC Press

1598_book.fm Page 316 Monday, October 11, 2004 1:57 PM

316

Signal and Image Processing in Navigational Systems

those in Equation (8.3) and Equation (8.6), deﬁne transformed target return signal ﬂuctuations and depend on time. This fact indicates a nonstationary state of target return signal ﬂuctuations.6–8 Dependence of the instantaneous power spectral densities of pulsed target return signal ﬂuctuations on time allows us to isolate information regarding pulsed target return signal delay, which deﬁnes the distance between the radar and the scanned target. As is well known, in the use of radar with frequencymodulated searching signals, information about the distance between the radar and the two-dimensional (surface) target is extracted from the frequency of the transformed target return signal. To obtain the frequency of the transformed target return signal from different targets, we use a set of ﬁlters. The bandwidth of each ﬁlter is approximately equal to the modulation frequency ΩM. The comb characteristics of these ﬁlters are the analogs of pulses gated in time under the use of radar with pulsed searching signals. For this reason, the correlation function of transformed target return signal ﬂuctuations is of prime interest to us, not the instantaneous correlation function. Thus, the correlation function of transformed target return signal ﬂuctuations must be averaged over the time period of frequency modulation.9,10 In principle, the correlation function of transformed target return signal ﬂuctuations averaged over time can be obtained based on Equation (8.10) and Equation (8.11), averaging over the period TM of the predetermined correlation function R(t, τ). Here we use the technique discussed in Section 2.3.4, which does not require determination of the instantaneous correlation function of the ﬂuctuations. For this purpose, each signal component given by Equation (8.8) and Equation (8.9) 2ρ c

e j[Ψ(t −0.5 τ ) − Ψ(t −0.5 τ −

∆ρ

+ c

)]

(8.12)

and

e − j[Ψ(t +0.5 τ ) − Ψ(t +0.5 τ −

2ρ c

∆ρ

− c

)]

(8.13)

is replaced in the integrand of Equation (8.10) with the equivalent Fourier series. In the integrand, we average the product of the Fourier series over the period TM of frequency modulation

R(τ) = p0 ⋅ e jωimτ

∫∫

g 2 (ϕ, ψ ) ⋅ e

− 2 jω av

∆ρ c

∞

⋅

∑C

n ρ− 0.5 ∆ρ

Cρn+0.5 ∆ρ ⋅ e

− jnω M ( τ −

∆ρ c

)

dϕ dψ ,

n= − ∞

(8.14) n

where Cρ−0.5 ∆ρ and Cρn+0.5 ∆ρ are the coefﬁcients of the Fourier-series expansion of signal components given by Equation (8.12) and Equation (8.13), and ωav is the high frequency averaged over the modulation period. Copyright 2005 by CRC Press

1598_book.fm Page 317 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

317

In an analogous manner, we can deﬁne the correlation function R(τ) if the searching signal has the form given by Equation (8.5). Unlike in Equation (8.10), differences between the function Ψ(t) and the coefﬁcients Cρn−0.5 ∆ρ and Cρn+0.5 ∆ρ of the Fourier-series expansion, respectively, should be estimated within the limits of several periods, as the arguments of the function Ψ(t) in Equation (8.8) and Equation (8.9) are shifted in each exponent. The correlation function averaged over time is the sum of the correlation functions given by Equation (8.14) with coefﬁcients Cρn±0.5 ∆ρ for each period. An example of a speciﬁc determination of R(τ) is discussed in Section 8.3. It should be noted that the general formula in Equation (8.14) coincides with the formula discussed in Jukovsky et al.,11 which is based on solving the electromagnetic problem of back scattering of radio waves by a rough surface under Gaussian statistics. Let us deﬁne the correlation function of the target return signal ﬂuctuations for the main forms of frequency-modulated searching signals.

8.2

The Linear Frequency-Modulated Searching Signal

Let us write the functions Ψ(t) and ω(t) in the following form:

Ψ(t) = 0.5kω t 2 and ω(t) = ω 0 + kω t ,

(8.15)

where kω is the velocity of frequency variation, which can be both positive (see Figure 8.1a) and negative (see Figure 8.1b). Assume that a linear variation in frequency is unlimited in time. In the case of the periodic searching signal, this really corresponds to the condition that the period of frequency modulation is very large in value so that Td << TM and τc << TM , where τc is the correlation length of the stochastic process. The difference ∆Ψ of the modulating functions given by Equation (8.3) can be written for the case of Equation (8.15) in the following form:

∆Ψ = kω (τ − 2 ∆ρc −1 )(t − 2ρc −1 ) .

(8.16)

Substituting Equation (8.15) and Equation (8.16) in Equation (8.3), we can write the correlation function R(t, τ) in the following form:

R(t, τ) = p0 ⋅ e jω (t )τ

∫∫

g 2 (ϕ, ψ ) ⋅ e

− 2 jω ( t )

∆ρ c

⋅e

− jkω ( τ −

2 ∆ρ 2 ρ c c

)

dϕ dψ .

(8.17)

Reference to Equation (8.17) shows that the target return signal is a nonstationary frequency-modulated stochastic process (the cofactor ejω(t)τ). The Doppler frequency is a function of time, too (the cofactor e Copyright 2005 by CRC Press

∆ρ

− 2 jω ( t ) c

). In addition

1598_book.fm Page 318 Monday, October 11, 2004 1:57 PM

318

Signal and Image Processing in Navigational Systems ω ωr (t ), V ≠ 0 ωr (t ), V = 0

Ω′0 (t ) − Ω′ρ

Ω′0(t ) Ωρ

ω (t ) = ω0 + k ω t, k ω > 0 ω0

t1

t

t1

t

(a) Td =

2ρ c

ω0

Ω′0(t ) Ω′0 (t ) + Ω′ρ

Ω′ρ

ωr (t ), V ≠ 0 ω (t ) = ω 0 − k ω t, k ω < 0 ωr (t ), V = 0 ω (b) FIGURE 8.1 Variation of instantaneous frequency of searching and target return signals under saw-tooth frequency modulation when the radar is stationary and is moving: (a) kω > 0 and (b) kω < 0.

Copyright 2005 by CRC Press

1598_book.fm Page 319 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

319

to the Doppler shift in frequency, the target return signal is shifted in frequency by the so-called range-ﬁnder frequency Ωρ (the cofactor e − 2 jk determined by

ω τρc

Ω ρ = kω 2ρc −1 .

−1

)

(8.18)

The range-ﬁnder frequency Ωρ is shifted by the Doppler frequency (the 2 jΩ

∆ρ

cofactor e ρ c ). Using Equation (8.10), we can deﬁne the correlation function of transformed target return signal ﬂuctuations, which is different from the correlation function given by Equation (8.17) only in that the cofactor ejω(t)τ is replaced with the cofactor ejωimτ, i.e., frequency modulation is absent in the case of high frequency. Consider slope scanning of the two-dimensional (surface) target when the velocity vector of moving radar is outside the directional diagram; the directional diagram is Gaussian; and the speciﬁc effective scattering area S°(ψ) is approximated by the exponent [see Equation (2.150)]. Using Equation (2.142), Equation (4.15), Equation (4.16), and Equation (8.17), we can deﬁne the instantaneous power spectral density of the ﬂuctuations in the following form: 2

( t )− Ω ′ ( t )] p(t) − π [ω −ω∆Ω S(ω , t) = ⋅e , ∆Ω FM 2 FM

(8.19)

where p(t) is the power of the target return signal in the absence of frequency modulation [see Equation (2.167)];

Ω ′(t) = Ω ′0 (t) − Ω ′ρ ; Ω ′0 (t) = Ω 0 (t) ⋅ (1 − δ 1 − δ 2 ) ; Ω 0 (t) = Ω max (t) ⋅ cos θ 0 ; Ω max (t) = 2Vc −1ω(t) ; Ω ′ρ = ω ρ0 [1 + 2Vc −1 cos θ 0 − δ 3 ] ≈ Ω ρ0 (1 − δ 3 ) ; Ω ρ0 = k ω ⋅

δ1 = −

2h = 2kω ρ0 c −1 ; c sin γ 0

k1 ∆(v2 ) a1 ; δ 2 = − 0.25π −1 ∆(v2 ) a1ctg γ 0 ; 4π cos θ0

δ 3 = 0.25π −1 ∆(v2 ) (k1 + ctg γ 0 ) ctg γ 0 ;

Copyright 2005 by CRC Press

(8.20)

(8.21)

1598_book.fm Page 320 Monday, October 11, 2004 1:57 PM

320

Signal and Image Processing in Navigational Systems ∆Ω FM = ∆Ω 2h + ( ∆Ω v + ∆Ω ρ )2 ; ∆Ω h = ∆(h2 )b1Ω max (t) ; ∆Ω v = ∆ a Ω max (t) ; ∆Ω ρ = ∆ Ω ρ0 ctg γ 0 ; (2) v 1

(8.22)

(2) v

cos θ 0 = cos ε 0 cos γ 0 cos(α + β 0 ) − sin ε 0 sin γ 0 ;

(8.23)

a1 and b1 are determined by Equation (II.26) (see Appendix II). As follows from Equation (8.19), the shape of the instantaneous power spectral density of the transformed ﬂuctuations at a ﬁxed t coincides with the shape of the power spectral density of the ﬂuctuations with the nonmodulated (in frequency) searching signal. The values of Ω′(t) and ∆ΩFM depend on the time t, too. The variation of the Doppler frequency Ω0(t) as a function of time is deﬁned by the values of kω and t. For instance, at f0 = 10 GHz, F0 = 104 Hz, and kf = 2000 GHz/sec for t1 = 10–2 sec, the change in the Doppler frequency Ω0(t) is 20 Hz. The frequency ωr(t) of the target return signal is plotted in Figure 8.1a and Figure 8.1b as a function of time at V = 0 and V ≠ 0. The frequencies Ω0′ (t) and Ωρ′ (t) are subtracted (kω > 0, V > 0, see Figure 8.1a) or added (kω < 0, V > 0, see Figure 8.1b), depending on the sign of kω. In addition, the instantaneous Doppler frequency as a function of carrier frequency at V ≠ 0 is shown in Figure 8.1. Under these conditions and with T

the ratio td10 = 0.5 × 10–3, the range-ﬁnder frequency Ωρ given by Equation (8.20) is equal to 10 kHz. The Doppler shift Ωρ′ in the range-ﬁnder frequency given by Equation (8.20) is equal to 0.01 Hz under the same conditions, i.e., it is inﬁnitesimal. The shifts δ1 and δ2 are the same as the shifts in the case of the nonmodulated (in frequency) searching signal with the speciﬁc effective scattering area S°(γ) and distance ρ. The shift δ3 in the range-ﬁnder frequency given by Equation (8.21) coincides with the shift δy1 + δρy [see Equation (II.19) and Equation (II.20), Appendix II] due to the speciﬁc effective scattering area S°(γ) and is equal to a timeshift of the pulsed target return signal center [see Equation (2.151)]. The bandwidth ∆ΩFM given by Equation (8.22) depends on time, too. However, this dependence is different for ∆Ωv , ∆Ωh , and ∆Ωρ. The values of ∆Ωh and ∆Ωv are independent of frequency [see Equation (3.76)], whereas the value ∆Ω

of ∆Ωρ depends on frequency due to ∆v . At Ωρρ = 0.1 and with the data given previously, the change in ∆Ωρ is equal to 2 Hz, i.e., we can neglect this value. Because of this, we can neglect changes in the power spectral density bandwidth of the ﬂuctuations caused by the function ω(t), the only exceptions being speciﬁc cases. Reference to Equation (8.22) shows that under the condition α = β0 = ε0 = 0, we can write

∆Ω FM = ∆(v2 ) [Ω ρctg γ 0 − 2Vc −1ω(t)sin γ 0 ] . Copyright 2005 by CRC Press

(8.24)

1598_book.fm Page 321 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

321

As one can see from Equation (8.24), at t = 0 and if the conditions

∆Ω v = ∆Ω ρ

Ω 0 = Ω ρ ctg 2 γ 0

and

(8.25)

are satisﬁed, the bandwidth ∆ΩFM is equal to zero. For the considered case, the value of ∆ΩFM varies from 0 at t = 0 to 2 Hz at t1 = 10–2 sec. In accordance with Equation (8.20), the values of Ω0 and Ωρ have different signs. Reference to Equation (8.25) shows that with power spectral density compression of the ﬂuctuations, we can write

V kω cos γ 0 = ⋅ . h f0 sin 3 γ 0

(8.26)

Consequently, varying the value of kω with known values of f0 and γ0 so that the bandwidth becomes minimum, we can deﬁne a navigational parameter ratio Vh . The effect of power spectral density compression of the ﬂuctuations can be explained based on the physical meaning, using Figure 8.2. Doppler Ω0 = const and range-ﬁnder Ωρ = const isofrequency lines in the direction of moving radar touch one another if the condition β0 = 0° is satisﬁed. With a decrease in the angle γ0 by 2∆V2 , the Doppler frequency increases by 0.5∆Ω, and vice versa. In accordance with Equation (8.20), the range-ﬁnder frequency must be negative and is equal to –Ωρ at the angle γ0. However, it is 0

equal to –Ωρ – 0.5∆Ωρ at the angle γ0 + 0

∆V 2 2

; and at the angle γ0 − 2∆V2 , we

obtain –Ωρ + 0.5∆Ωρ. Consequently, at the angle γ 0 + 0

∆V 2 2

, the sum of the

range-ﬁnder frequency shift and the Doppler shift is given by

Ω 0 + 0.5∆Ω − Ωρ0 − 0.5∆Ω ρ = Ω 0 − Ω ρ0 ;

(8.27)

at the angle γ0 − 2∆V2 , we have

Ω 0 + 0.5∆Ω − Ωρ0 + 0.5∆Ω ρ = Ω 0 − Ω ρ0 ;

(8.28)

and at the angle γ0, we obtain Ω0 – Ωρ0. Thus, relative shifts in the radar-range frequency Ωρ and Doppler 0 frequency Ω0 with respect to the difference Ω0 – Ωρ within the directional 0 diagram compensate each other at the angle γ0. If the Doppler frequency Ω0 is negative, the range-ﬁnder frequency Ωρ has to be positive to ensure power 0 spectral density compression of the ﬂuctuations under sloping scanning of the underlying surface of a two-dimensional target. It should be noted that complete compression can take place not only in the horizontal direction of

Copyright 2005 by CRC Press

1598_book.fm Page 322 Monday, October 11, 2004 1:57 PM

322

Signal and Image Processing in Navigational Systems Y γ0 −

∆ν 23/2 γ0 γ0 +

Vertical way Ω0

∆ν 23/2 Ω0 − 0.5∆ Ω Horizontal way Ω0

0

Ω 0 + 0.5∆Ω

− Ωρ

0

Z

X

− Ωρ − 0.5∆ Ωρ 0

− Ωρ + 0.5∆Ωρ 0

FIGURE 8.2 The Ω0 and Ωρ isofrequency lines on the scattering surface.

moving radar, but in all cases when the velocity vector of moving radar and the directional diagram axis are in the same vertical plane. However, it should be noted that the average [see Equation (8.19)] power spectral density frequency of the ﬂuctuations varies in time according to the saw-tooth law due to variation of the functions ω(t) and Ω′(t) both in the presence and absence of compression. The Doppler frequency as a function of time is present in the transformed target return signal and can hide the effect of compression.12,13 Using Equation (2.66) and Equation (8.19), we can easily deﬁne the power spectral density S(ω ) of the nonperiodic ﬂuctuations averaged over the time interval [0, t1]. If we do not take into consideration the effect of the term kωt on the value of ∆ΩFM [see Equation (8.22)], the power spectral density S(ω , t) of the ﬂuctuations averaged over time takes the following form:

{

S(ω , t) = S(ω) = 0.5p(t) ⋅ Φ [ ∆ΩπFM (ω − Ω′0 − ω 0 + Ωρ′ + 0.5∆ω M )] – Φ

[

π ∆Ω FM

]}

(ω − Ω′0 − ω 0 + Ωρ′ − 0.5∆ω M ) , (8.29)

Copyright 2005 by CRC Press

1598_book.fm Page 323 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

323

S (ω, t )

3

3

2

2 1

ω ω0 + Ω′0 − Ωρ

ω0 + Ω ′0 − Ωρ + ∆ω2

ω 0 + Ω 0′ − Ωρ + ∆ω 3

FIGURE 8.3 The power spectral density (averaged over time) of the saw-tooth frequency-modulated target return signal: (1) ∆ω1 = 0; (2) ∆ω2 ≠ 0, ∆ω2 < ∆ω3; (3) ∆ω3 ≠ 0, ∆ω2 < ∆ω3.

where ∆ωM = Kωt1 and Φ(x) is given by Equation (1.27). The power spectral density of the ﬂuctuations given by Equation (8.29) is shown in Figure 8.3. The power spectral density bandwidth at the level 0.5 is equal to the deviation of the frequency ∆ωM; the width of front and cut is equal to ∆ΩFM. Omitting the intermediate mathematics, we consider the power spectral density of the ﬂuctuations for the case when the directional diagram axis is directed vertically down and the velocity vector of moving radar is located on this axis. We assume that the directional diagram is axial symmetric and has width ∆; and the speciﬁc effective scattering area S°(ψ) is approximated by the Gaussian law [see Equation (2.187) and Equation (II.8), Appendix II]. Using the same technique as before, we obtain

S(ω) =

p(t) − π ω − ω∆Ω( t ) − Ω( t ) ⋅e , ∆Ω FM FM

(8.30)

where ω < Ω(t) + ω(t) at Ω0 ≥ Ωρ and ω > Ω(t) + ω(t) at Ω0 ≤ Ωρ ; p(t) is the 0 0 target return signal power given by Equation (2.203);

Ω(t) = Ω 0 (t) − ∆Ω ρ0 ; Ω 0 (t) = − ε 0 = ±90 ; ∆Ω FM

2V ⋅ ω(t)sin ε 0 , c

∆2 | Ω 0 (t) + Ω ρ0 | 2h ; ∆Ω ρ0 = = ⋅ kω . 4 πa 2 c

(8.31)

The power spectral density of the ﬂuctuations has an exponential shape, as in the case of the non-frequency-modulated searching signal. The bandwidth Copyright 2005 by CRC Press

1598_book.fm Page 324 Monday, October 11, 2004 1:57 PM

324

Signal and Image Processing in Navigational Systems

∆ΩFM of the power spectral density depends on the speciﬁc effective scattering area S°(ψ) (see in Section 4.7, the term a2 in the denominator) and on time, just as the average Doppler frequency given by Equation (8.31) depends on time. Reference to Equation (8.31) shows that if the condition t = 0 is satisﬁed and the following equality

2Vc −1ω 0 sin ε 0 = 2hc −1kω

(8.32)

is true, the power spectral density is compressed to the bandwidth deﬁned by the function ω(t) in Equation (8.31). In the present case, this can be explained by complete coincidence between the Doppler and range-ﬁnder isofrequency lines, which are circles. Unlike in sloping scanning, power spectral density compression takes place if the values of Ω0 and Ωρ have the 0 same sign, as the Doppler frequency Ω0 increases with deﬂection of the directional diagram axis from ε0 = 90°.

8.3

The Asymmetric Saw-Tooth Frequency-Modulated Searching Signal

In this case, the time-periodic sequence of searching signals coherent from period to period is given by Equation (8.5), where the functions Ψ(t) and ∆ω ω(t) are determined by Equation (8.15) and kω = TMM . Changes in the frequency of this searching signal are shown in Figure 8.4. Figure 8.5 shows the frequency ωh of the heterodyne signal and the frequency ωr of the target return signal as a function of the distance ρ. It is assumed in Figure 8.5 that

ω im = 0 and V = 0 ; t0 = 0.5τ ,

t1 = 0.5τ + 2ρc −1 ,

and t3 = 0.5τ + TM + 2ρc −1 .

t2 = 0.5τ + TM ,

(8.33)

(8.34)

The transformed target return signal Wr(t + 0.5τ) shifted left by τ with respect to the signal Wr(t – 0.5τ) on the time axis presents an analogous picture. In the general case, the signals Wr(t – 0.5τ) and Wr(t + 0.5τ) may not overlap within the limits of the period n = 0. Reference to Figure 8.6 shows that the signal Wr(t – 0.5τ) or Wr(t + 0.5τ) consists of two closing pulses within the limits of each period TM: the ﬁrst pulse with frequency kωTd and duration TM – Td and the second pulse with frequency kω(TM – Td) and duration Td . For

Copyright 2005 by CRC Press

1598_book.fm Page 325 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

325

ω

∆ω M

ω0 −2TM

−TM

t 0

TM

2TM

FIGURE 8.4 The frequency of the searching signal under asymmetric saw-tooth frequency modulation.

ω

2

1

ω0 0

t0

t1

t2

t3

t

FIGURE 8.5 The frequency of the heterodyne (1) and target return (2) signals under asymmetric saw-tooth frequency modulation.

further analysis, deﬁne differences ∆Ψ1(t) in the exponents of Equation (8.10), which depend on t – 0.5τ and t + 0.5τ:

∆Ψ1 (t ∓ 0.5τ) = ± kω (t ∓ 0.5τ) ⋅ ( 2cρ ∓

∆ρ c

).

(8.35)

We can deﬁne the functions ∆Ψ1(t) within the limits of the period TM. For all z1,2 = t ∓ 0.5τ we can use the Fourier-series expansion Copyright 2005 by CRC Press

1598_book.fm Page 326 Monday, October 11, 2004 1:57 PM

326

Signal and Image Processing in Navigational Systems ω h − ωr

k ωTd

t

0

k ω(TM − Td )

FIGURE 8.6 The frequency of the transformed target return signal under asymmetric saw-tooth frequency modulation. ∞

Per TM (e

jkω z1Td′

)= ∑[ n= − ∞

∞

Per TM (e

jkω z2Td′′ ∗

)

=

1 TM

TM

∫e

n= − ∞

⋅ e − jnω

M z1

dz1 ⋅ e jnω

jkω z2Td′′

⋅ e − jnω

M z2

dz2 ⋅ e − jnω

M z1

;

(8.36)

0

TM

∑[ ∫ e 1 TM

]

jkω z1Td′

]

M z2

,

(8.37)

0

where PerTM(x) denotes the periodicity of the function with period TM;

Td′ = 2(ρ − 0.5∆ρ)c −1 and Td′′= 2(ρ + 0.5∆ρ)c −1 ;

(8.38)

ωM is the modulation frequency. In accordance with Figure 8.4–Figure 8.6, the coefﬁcients of the Fourier-series expansion in Equation (8.36) and Equation (8.37) should be deﬁned for the function given by Equation (8.36) within the limits of the intervals [t1, t2] and [t2, t3] independently of the function given by Equation (8.37). This is also true for the function given by Equation (8.37). It is necessary to substitute Equation (8.36) and Equation (8.37) in Equation (8.10) and to average over time within the limits of the period TM. Carrying out the mathematics, we can deﬁne the correlation function of the ﬂuctuations averaged over time for arbitrary positions of the velocity vector of moving radar and radar antenna beam with respect to the twodimensional (surface) target

Copyright 2005 by CRC Press

1598_book.fm Page 327 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

R(τ) = p0 ⋅ e jωimτ ∞

×

∑

n= − ∞

+

Td′ TM

∫∫

g 2 (ϕ, ψ ) ⋅ e

∆ρ c

−2 j ( ω 0 + 0.5 ∆ω M )

[(1 − )(1 − ) ⋅ sinc x′ sinc x′′ Td′ TM

Td′′ TM

⋅ TTMd′′ ⋅ sinc y′ sinc y′′ ] ⋅ e

− jnω M ( τ −

327

∆ρ c

)

,

(8.39)

dϕ dψ

where

x ′ = 0.5(kω Td′ − nω M ) ⋅ (1 − TTMd′ ) ⋅ TM and x ′′ = 0.5(kω Td′′− nω M ) ⋅ (1 −

Td′′ TM

) ⋅ TM ;

y ′ = 0.5 ⋅ [kω (TM − Td′) + nω M ] ⋅ T ′ and y ′′ = 0.5 ⋅ [kω (TM − Td′′) + nω M ] ⋅ Td′′; sinc x =

sin x . x

(8.40)

(8.41)

(8.42)

Comparing Equation (8.14) and Equation (8.39), one can see that the coefﬁcients Cρ″ have the form of the function sinc x and ωav = ω0 + 0.5∆ωM. The power spectral density of the ﬂuctuations consists of a set of partial power spectral densities deﬁned by harmonics of the modulation frequency ωM . Unlike the case of the nonperiodic searching signal given by Equation (8.17), the correlation function given by Equation (8.39) possesses two range-ﬁnder frequencies

Ω ′ρ = kω Td′ and Ω ρ′′ = kω (TM − Td′)

(8.43)

(see Figure 8.6), which are deﬁned by different intervals of the modulation period. Harmonics of the modulation frequency ωM are shifted by the oneω ∆ρ half Doppler frequency Mc . Because all components of the target return signal should be shifted by their Doppler frequency, not the one-half Doppler frequency, we can reason that another one-half Doppler shift in frequency is given by Equation (8.39) using the arguments kωTd′ and kωTd″ of the sinc x functions. As the condition nωM << ω0 is satisﬁed, as a rule, we can neglect the Doppler shift in the frequency ωM. Reference to Equation (8.39) shows that the target return signal power at the range-ﬁnder frequency kωTd is proportional to

Copyright 2005 by CRC Press

1598_book.fm Page 328 Monday, October 11, 2004 1:57 PM

328

Signal and Image Processing in Navigational Systems

(1 − TT ′ ) ⋅ (1 − TT ′′ ) ≈ (1 − TT )2 , d

d

d0

M

M

M

(8.44)

i.e., to the square of the relative decrease in the period TM , as we can suppose here that

Td′ = Td′′= Td0 = 2ρ0 c −1 ,

(8.45)

where ρ0 corresponds to the distance in the direction along the directional diagram axis. The second term in Equation (8.39) deﬁnes the part of the power spectral density that lies within another modulation period. The power of this part is proportional to 2 Td′ Td′′ Td0 ⋅ ≈ 2 . TM TM TM

(8.46)

If the ﬁrst target return signal with the range-ﬁnder frequency kωTd0 becomes maximum at n0ωM, the second target return signal with the range-ﬁnder frequency kω(TM – Td0) becomes maximum under the condition

kω (TM − Td0 ) = −n1ω M .

(8.47)

If the condition TM >> Td0 is satisﬁed, we can write

kω (TM − Td0 ) ≈ ∆ω M ,

(8.48)

i.e., the frequency given by Equation (8.48) is approximately equal to the frequency deviation. The maximum of the power spectral density of the ﬂuctuations of the ﬁrst target return signal takes place at the frequency ωim + Ω0 – n0ωM. The maximum of the power spectral density of the ﬂuctuations of the second target return signal takes place at the frequency ωim + Ω0 + n1ωM.The power ratio of these components is equal to

2 TM

Td2

and is independent

0

of the value of ∆ωM. For instance, at ρ0 = 3 km, Td0 = 2 · 10–5 sec, and TM = 2 × 10–2 sec, we get

TM Td

= 103. For this reason, here, we can neglect the second

0

term. If the condition TM < Td0 is satisﬁed, Equation (8.39) is true, too. Then, the parameter Td must be expressed in the form lTM + td , where td ≤ TM , and we need to replace the parameter Td with the parameter td in Equation (8.39). The frequencies kωTd0 and kω(TM – Td0), as in the case of radar with pulsed searching signals, do not contain any information regarding lTM if the period

Copyright 2005 by CRC Press

1598_book.fm Page 329 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

329

is less than Td and will be proportional to kωtd and kω(TM – td), respectively. Henceforth, we assume that Td′ << TM , Td″ << TM , and Td′ = Td″ = Td , as we can neglect the effect of the parameter ∆ρ on the parameters Td′ and Td″ . In this case, we can write

x ′ = x ′′ =

π(kω Td − nω M ) = π( ∆fM Td − n) . ωM

(8.49)

Denote

Xρ = ∆fM Td = 2∆fM

ρ Ωρ = . c ωM

(8.50)

The pure parameter Xρ is normalized with respect to the modulation frequency ωM by the range-ﬁnder frequency Ωρ. The sign of the parameter Xρ depends on the sign of the parameter ∆fM. Because

ρ=

h , sin(ψ + γ 0 )

(8.51)

the pure parameter Xρ can be thought of as a function of the angle ψ. For simplicity and convenience, in the subsequent discussion, we will use the frequency ω0 instead of the frequency ω0 + 0.5∆ωM. Taking into consideration the previously mentioned statements, the correlation function of the ﬂuctuations given by Equation (8.39) can be written in the following form:

R(τ) = p0 ⋅ e jωimτ

∫∫

g 2 (ϕ, ψ ) ⋅ e

−2 jω 0

∆ρ c

∞

∑ sinc [π(X 2

ρ

− n)] dϕ dψ . (8.52)

n= − ∞

Let us consider some speciﬁc cases.

8.3.1

Sloping Scanning

As follows from Section 8.2, the power spectral density characteristics of the target return signal ﬂuctuations with the range-ﬁnder frequency depend only on characteristics of the vertical-coverage directional diagram. Because of this, consider ﬁrst the case of sloping scanning (i.e., the condition γ0 + ε0 < 90° should be satisﬁed), when the velocity vector of moving radar lies in the same vertical plane as the vertical-coverage directional diagram axis, is outside the directional diagram, and is toward directed, i.e., α = β0 = 0°. ~ Let us assume that the variables ϕ and ψ in the function g(ϕ, ψ) are separable and that the speciﬁc effective scattering area S° is a function of the Copyright 2005 by CRC Press

1598_book.fm Page 330 Monday, October 11, 2004 1:57 PM

330

Signal and Image Processing in Navigational Systems

angle γ only. Under these conditions, Equation (8.50) can be written in the following form:

Xρ = Xρ0 (1 − ψ ctg γ 0 ) and Xρ0 = 2∆fM hc −1 sin γ 0 = ∆fM Td0 . (8.53) Using the Fourier transform for Equation (8.52) and taking into consideration the dependence between Xρ and ψ, we can write the power spectral density of the ﬂuctuations in the following form: ∞

S(ω) =

∑

p1 − Ωn ⋅ g 2 [ ΩΩn −v ω ] ⋅ sinc 2 π[Xρ0 − n + ω∆Ω ⋅ dρ ] , v ∆Ωv n = − ∞

(8.54)

where

Ω n = ω im + Ω 0 − nω M ; dρ =

(8.55)

∆Ω ρ = ∆fM Tr = Xρ0 ∆(v2 )ctg γ 0 ; ωM

(8.56)

the parameters p1, Ω0, Ωv , and ∆Ωv are the same as in the absence of frequency modulation:

Ω 0 = 4 π Vc −1 cos(γ 0 + ε 0 ) ; Ω v = 4π Vc −1 sin(γ 0 + ε 0 ) ; and ∆Ω v = Ω v ∆(v2 ) ;

p1 =

∫

(8.57)

PG02 λ 2 ∆(v2 )S°(γ 0 )sin γ 0 gh2 (ϕ) dϕ 64π 3 h2

.

(8.58)

The parameter dρ is the bandwidth of the range-ﬁnder spectrum ∆Ωρ normalized with respect to the modulation frequency ωM [see Equation (8.22)]. If the condition ∆fM = 0 is satisﬁed, the power spectral density given by Equation (8.54) deﬁnes the power spectral density of the target return signal Doppler ﬂuctuations in the absence of frequency modulation, which is shifted by the frequency ωim:

S(ω) =

Copyright 2005 by CRC Press

p1 ⋅ g 2 ( Ω0 +Ωωvim − ω ) . ∆Ωv v

(8.59)

1598_book.fm Page 331 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

331

Assume that the directional diagram is Gaussian and that the speciﬁc effective scattering area S°(ψ) is deﬁned by exponent.14,15 As is well known, the main lobe of the function sinc2 x can be approximated by the Gaussian law:

sinc 2 π(Xρ − n) = e

− π ( Xρ − n )2

.

(8.60)

Using the approximation given by Equation (8.60), both functions have the same effective bandwidth 1 (or ωM in the nonnormalized case). In this case, based on Equation (8.54) we can write ∞

−π pFM S(ω) = ⋅ e | ∆ΩvFM| n = − ∞

∑

( Xρ′ − n )2

−π

0

⋅e

1+ dρ2

( ω − Ωn )2 ∆Ω2

v FM

,

(8.61)

where

pFM =

PG02 λ 2S°(γ 0 )∆ v ∆ h sin γ 0 128π h 1 + d 3

Ωn = ω im + Ω′0 − nω M −

2

2 ρ

∆(v2 ) ( k1 + ctg γ 0 )

⋅e

8 π ( 1 + dρ2 )

;

∆Ωv dρ (Xρ0 − n) 1 + dρ2

∆Ωv dρXρ0 ∆Ωv dρ = ω im + Ω′0 − nω M − +n − ωM 2 1 + dρ 1 + dρ2

[

;

(8.63)

]

Ω ′0 = Ω 0 (1 − δ 1FM − δ 2FM ) ;

(8.64)

k1∆(v2 ) tg ( γ 0 + ε 0 ) ; 4 π(1 + dρ2 )

(8.65)

∆(v2 )ctg γ 0 ⋅ tg ( γ 0 + ε 0 ) ; 4 π(1 + dρ2 )

(8.66)

δ 1FM =

δ 2FM =

∆Ω vFM =

Copyright 2005 by CRC Press

(8.62)

∆Ω v

;

(8.67)

Xρ′0 = Xρ0 (1 − δ 3 ) .

(8.68)

1 + dρ2

1598_book.fm Page 332 Monday, October 11, 2004 1:57 PM

332

Signal and Image Processing in Navigational Systems

Reference to Equation (8.62), Equation (8.64)–Equation (8.66), and Equation (8.67) shows that the parameters pFM, δ1FM, δ2FM, and ∆ΩvFM are equal to the corresponding parameters of the power spectral density of the ﬂuctuations in the absence of frequency modulation, replacing the parameter ∆v with the parameter

∆v 1 + dρ2

. The shift δ3 in the normalized range-ﬁnder frequency [see

Equation (8.68)] is equal to the range-ﬁnder frequency shift of the nontransformed target return signal [see Equation (8.21)]. First, consider the stationary radar, i.e., the condition V = 0 is true. In this case, under the condition ∆ΩvFM → 0, based on Equation (8.61) we can determine the power spectral density of the ﬂuctuations as follows: ∞

S(ω) = pFM

∑

δ[ω − (ω im − nω M )] ⋅ e

−π

( Xρ′ − n )2 0

1 + dρ2

.

(8.69)

n= − ∞

Reference to Equation (8.69) shows that the power spectral density is a set of discrete components at the frequencies ωim – nωM. The power of each component is deﬁned by the modulating function given by Equation (8.60) and by the parameter dρ, which depends on the parameters of the verticalcoverage directional diagram. Under the condition Xρ′ > 0, there are only 0 harmonics at frequencies ωim – nωM in the power spectral density and under the condition Xρ′ < 0, only harmonics at frequencies ωim + nωM. If the condi0 tions dρ << 1 and Ωρ < ωM are satisﬁed, based on Equation (8.69) we can write 0

∞

S(ω) = pFM

∑ δ[ω − (ω

im

− nω M )] ⋅ e

− π ( Xρ′ 0 − n)2

(8.70)

n= − ∞

and the power spectral density is independent of the directional diagram and is deﬁned only by the modulating function given by Equation (8.60). The effective bandwidth of this power spectral density envelope is equal to 1 or ωM for nonnormalized values. The only harmonic n0ωM takes place under the condition Xρ′ = n0 within the limits of the effective power spectral density 0 bandwidth given by Equation (8.70). In spite of the fact that in the present case, the target return signal consists of the individual target return signals from a set of random scatterers spaced by the distance ρ0, the shape of the power spectral density of the ﬂuctuations corresponds to reﬂections from the point target. As can be easily seen, the condition dρ << 1 corresponds to the condition δρ0 >> δρ, where

δρ0 =

Copyright 2005 by CRC Press

c 2∆fM

and

δρ =

h∆ v ctg γ 0 . 2 sin γ 0

(8.71)

1598_book.fm Page 333 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

333

As is well known, the parameter δρ0 is the radar resolution with the frequency-modulated searching signal for point targets; δρ is the length of the scanned area of the two-dimensional (surface) target (see Figure 8.7 and Figure 8.8). The radar resolution with the frequency-modulated searching signal depends only on frequency deviation and corresponds to the resolution 0.5cτp of radar with the pulsed searching signal under the condition τp = ∆f1M . If the condition δρ0 >> δρ is satisﬁed, we can say that reﬂection is caused by an area no larger than one zone of the radar range (see Figure 8.7). Under the condition dρ >> 1, we can assume that Ωρ >> ωM and the power 0 spectral density takes the following form: ∞

S(ω) = pFM

∑ δ[ω − (ω

im

− nω M )] ⋅ e

−π

( Xρ′ − n )2 0

dρ2

.

(8.72)

n= − ∞

In this case, the power spectral density envelope of the ﬂuctuations in nonnormalized units with respect to the modulation frequency ωM coincides with the power spectral density envelope of the ﬂuctuations with the rangeﬁnder frequency, the bandwidth of which is equal to ∆Ωρ , and is deﬁned completely by the shape of the vertical-coverage directional diagram. Under V Y ε0 γ0

ρ2 δρ0 =

c 2∆fM

ρ1

X

n0 FIGURE 8.7 A single radar range zone.

Copyright 2005 by CRC Press

1598_book.fm Page 334 Monday, October 11, 2004 1:57 PM

334

Signal and Image Processing in Navigational Systems V Y ε0 γ0

ρ2

δρ

ρ1

X

n0 − k

δρ0 =

n0

n0 + k

c 2 ∆fM

FIGURE 8.8 Several radar range zones ( δρδρ0 = 11) .

the condition ωM → 0 or TM → ∞, the power spectral density given by Equation (8.72) corresponds to the power spectral density of the nonperiodic ﬂuctuations given by Equation (8.19) for the stationary radar. The condition dρ >> 1 is analogous to the condition δρ0 << δρ. For this case, the number of radar range zones is equal to dρ within the scanned area of the two-dimensional (surface) target (see Figure 8.8). Each harmonic ωim – (n ± k) × ωM is formed in its own scanned area, as shown in Figure 8.8. During formation, the neighboring harmonics are noncoherent relative to one another and are formed independently. Examples of power spectral densities in the case of stationary radar are shown in Figure 8.9 and Figure 8.10. Now, consider the power spectral density with moving radar, i.e., V ≠ 0.16 If the condition dρ << 1 is true, we can neglect the parameter dρ and the terms in which the parameter dρ is a cofactor, in Equation (8.54) and Equation (8.61). In the present case, under the condition Xρ′ = n0 , all partial power spectral 0 densities are inﬁnitesimal at n ≠ n0. Reference to Equation (8.54) and Equation (8.61) shows that

S(ω) =

Copyright 2005 by CRC Press

p

⋅ g 2 (

| ∆Ω v |

Ωn0 − ω Ωv

),

(8.73)

1598_book.fm Page 335 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

335

S 0 (ω) p

sinc2 [π(X ρ′ 0 − n)]

ω n0 + 3

n0 + 2

n0 + 1 ω im − n0 ωM n0 − 1

n0 − 2

n0 − 3

ω im

FIGURE 8.9 The power spectral density of target return signal ﬂuctuations with stationary radar and the asymmetric saw-tooth frequency-modulated searching signal: dρ << 1.

S 0 (ω) p

dρ

ω n0 + 3

n0 + 2

n0 + 1 ω im − n0 ω M n0 − 1

n0 − 2

n0 − 3

ω im

FIGURE 8.10 The power spectral density of target return signal ﬂuctuations with stationary radar and the asymmetric saw-tooth frequency-modulated searching signal: dρ >> 1.

Copyright 2005 by CRC Press

1598_book.fm Page 336 Monday, October 11, 2004 1:57 PM

336

Signal and Image Processing in Navigational Systems

where

Ω n0 = ω im + Ω ′0 − n0 ω M ,

(8.74)

i.e., only one power spectral density of the Doppler ﬂuctuations takes place at the frequency Ωn0. Under the condition dρ >> 1, the maximums of all partial power spectral densities of the ﬂuctuations are deﬁned by the “equality to zero” condition of the argument of the function sinc2 x in Equation (8.54). For instance, the maximum partial power spectral density corresponding to the (n0 + k)-th harmonic takes place at the frequency

ω k = Ω n0 + k + k ⋅

∆Ω v , dρ

(8.75)

where

Ω n0 + k = ω im + Ω ′0 − (n0 + k )ω M .

(8.76)

Based on the physical meaning we can explain this in the following manner. The vertical-coverage directional diagram is divided into dρ partial directional diagrams due to the frequency-modulated searching radar signal as shown in Figure 8.7 and Figure 8.8. The Doppler shift in frequency corresponding to the middle of the k-th radar range zone relative to n0 is equal Ω′ + k∆Ω

to 0 dρ v . As a consequence, the maximums of the partial power spectral densities of the ﬂuctuations do not take place at the frequencies given by k∆Ω

Equation (8.76), but are shifted by dρ v relative to these frequencies [see Equation (8.75)]. This follows from Equation (8.61), too, if we assume that

dρ2 ≈ 1 + dρ2 .

(8.77)

Ω −ω

Substituting ωk for ω in the function g˜ v2 ( Ωn v ) and taking into consideration the functional relationship between Ωv and ∆Ωv , the maximums of the partial power spectral densities of the ﬂuctuations (or their envelopes) are deﬁned as follows:

S(ω k ) =

(2) p1 ⋅ g˜ v2 (− k∆dρv ) . ∆Ω v

The shape of the partial power spectral density is determined by

Copyright 2005 by CRC Press

(8.78)

1598_book.fm Page 337 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

Sn0 + k (ω) =

The function sinc2

[

p1 ⋅ g 2 ∆Ωv v

π ( ω − ω k ) dρ ∆Ωv

(

ωk −

k ∆Ωv dρ

−ω

Ωv

) ⋅ sinc [

2 π ( ω − ω k ) dρ ∆Ωv

].

337

(8.79)

] will “cut” the power spectral density with an

effective bandwidth equal to

∆Ω v dρ

from the function g˜ v2 (ψ ) under tuning on

k∆Ω

the frequency dρ v . In the general case, the power spectral density is not symmetrical, because the “cutting” is carried out on the slope of the function

g˜ v2 (ψ ). Only in the case of the Gaussian vertical-coverage directional diagram and if the condition in Equation (8.60) is satisﬁed, as shown from Equation (8.61), is the power spectral density symmetric, with a bandwidth of

∆Ω v dρ

. Reference to Equation (8.75) shows that in accordance with the sign of the parameters ∆Ωv and dρ , the frequency ωk can be greater than the frequency Ωn +k (if ∆Ωv and dρ have the same sign) or less (if ∆Ωv and dρ have different 0 signs). In the present case, the bandwidth ∆Ωv is positive, i.e., the condition α = β0 = 0° is satisﬁed. Let us deﬁne a difference between the frequencies v ω k +1 − ω k = ω M ( d∆Ω − 1) . ρω M

(8.80)

One can see from Equation (8.80) that if the condition

∆Ω v = dρω M

(8.81)

is true, the partial power spectral densities are superimposed on one another. In this case, the power spectral density given by Equation (8.61) has the following form: −π p S(ω) = 1 ⋅ e ωM

( ω − Ω′n )2 0

ω 2M

,

(8.82)

where

Ω ′n0 = ω im + Ω ′0 − n0 ω M = ω im + Ω ′0 − Xρ′0 ω M .

(8.83)

The effective compressed power spectral density bandwidth corresponds to reﬂection from a single radar range zone [see Equation (8.71)] and is equal to ωM. To compress the power spectral density, it is necessary to satisfy the condition given by Equation (8.81), and the Doppler and range-ﬁnder frequencies [see Equation (8.83)] must have different signs.

Copyright 2005 by CRC Press

1598_book.fm Page 338 Monday, October 11, 2004 1:57 PM

338

Signal and Image Processing in Navigational Systems

If the condition α + β0 = π is true, i.e., if the radar antenna beam is directed back with respect to the line of radar motion, the parameter dρ must be negative to compress the power spectral density. In this case,

Ω ′n0 = ω im − Ω ′0 + n0 ω M .

(8.84)

The physical signiﬁcance of power spectral density compression is discussed in Section 8.2. However, in the present case, the searching signal is periodic, and power spectral density compression is possible only up to the modulation frequency ωM. In the case of the periodic searching signal, a complete power spectral density compression of the ﬂuctuations takes place only under the condition t = 0, and changes in time of the Doppler frequency are retained due to carrier frequency variation with the frequency-modulated searching radar signal. This variation is averaged in the correlation function (averaged over time) of the transformed ﬂuctuations and a power spectral density compression of the ﬂuctuations takes place at the frequency ω0 + 0.5ωM. To deﬁne the power spectral density under the condition α + β0 ≠ 0° or π, it is necessary to carry out a convolution of the power spectral density given by Equation (8.61) and the power spectral density given by Equation (4.22) for the continuous nonmodulated searching signal or carry out the mathematics discussed in Section 8.2. Omitting the mathematics used in accordance with Baker,17 we can say that the power spectral density of the ﬂuctuations will coincide with the power spectral density given by Equation (8.61) on replacing the bandwidth ∆Ωv FM with the bandwidth ∆Ωv in Equation (8.61), where

∆Ω FM = ∆Ω 2vFM + ∆Ω 2h

and ∆Ω h = 4 2 π Vλ−1b1∆ h ,

(8.85)

and the frequencies Ω0 and Ωv with the frequencies

Ω 0 = 4 π Vλ−1 cos θ 0

(8.86)

Ωv = − 4π V λ −1 a1 ,

(8.87)

and

respectively. If the condition α + β0 = 0° is not satisﬁed, a compression of the bandwidth ∆Ωv FM is only possible for an arbitrary position of the sloped radar antenna beam. If the condition given by Equation (8.81) is satisﬁed, the power spectral density bandwidth is determined by

∆Ω vFM = ω 2M + ∆Ω 2h .

Copyright 2005 by CRC Press

(8.88)

1598_book.fm Page 339 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

339

The partial power spectral densities with the use of the Gaussian approximation and approximation by the function sinc2 x for the vertical-coverage directional diagram are shown in Figure 8.11 and Figure 8.12, respectively, at the same and different signs of the frequencies Ω0′ and nωM for the frequency Ωn. In the second case, as one can easily see, partial compression of the symmetric spectral density takes place. To deﬁne the total power spectral densities of target return signal ﬂuctuations shown in Figure 8.11 and Figure 8.12, it is necessary to add the ordinates of all partial power spectral densities. S (ω) Smax (ω) 1.00

0.75

0.50

0.25 n 0.00

171

173

175

177

179

181

183

185

FIGURE 8.11 Partial power spectral densities of target return signal ﬂuctuations with an asymmetric sawtooth frequency-modulated searching signal at the total frequency: dρ >> 1.

S (ω) Smax (ω) 1.00

0.75

0.50

0.25 n 0.00

85

87

89

91

93

95

97

99

FIGURE 8.12 Partial power spectral densities of target return signal ﬂuctuations with the asymmetric sawtooth frequency-modulated searching signal at the difference frequency: dρ >> 1.

Copyright 2005 by CRC Press

1598_book.fm Page 340 Monday, October 11, 2004 1:57 PM

340 8.3.2

Signal and Image Processing in Navigational Systems Vertical Scanning and Motion

Under vertical scanning, the condition γ0 = 90° is true. Assume that the velocity vector of moving radar is directed along the directional diagram axis and is equal to V · sin ε0.18,19 The Doppler frequency Ω0 is negative under the condition ε0 = 90° and is positive under the condition ε0 = –90°. Suppose directional diagram has axial symmetry and depends on both the speciﬁc effective scattering area S° and the angle in the vertical plane. Omitting the mathematics, we can write the power spectral density of target return signal ﬂuctuations in the following form: ∞

S(ω) =

2 π p0 ⋅ g 2 | Ω0| n = − ∞

∑ (

Ωn − ω 0.5 Ω0

)⋅ sinc {π [X 2

h

− n + (Ωn Ω− ω0 )Xh ]} , (8.89)

where

g 2

(

Ω n = ω im + Ω 0 − nω M ;

(8.90)

Ωn − ω ≥ 0; Ω0

(8.91)

X h = 2h∆fM c −1 ;

(8.92)

Ωn − ω 0.5 Ω0

)= g ( 2

Ωn − ω 0.5 Ω0

) ⋅ S°(

Ωn − ω 0.5 Ω0

).

(8.93)

The function g˜ 2 (ψ ), as in the case of sloping scanning, deﬁnes the shape of the power spectral density in the absence of frequency modulation. The parameter Xh , as in Section 8.3.1, is the range-ﬁnder frequency (normalized with respect to the modulation frequency ωM) corresponding to the direction of the directional diagram axis. Approximating the directional diagram and speciﬁc effective scattering area S° by the Gaussian law, based on Equation (8.89) the power spectral density has the form: ∞

S(ω) =

Ω −ω p0′ − ⋅ sinc 2 {π[Xh − n + (Ωn2−∆Ωω )dh ]} ⋅ e ∆Ω , |∆Ω| n = − ∞

∑

n

(8.94)

where

ω < Ω n at Ω 0 > 0, and

Copyright 2005 by CRC Press

ω > Ω n at Ω 0 < 0;

(8.95)

1598_book.fm Page 341 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

341

dh = 0.5π −1∆( 2 ) X h a −2 ; a 2 = 1 + 0.5π −1k2 ∆( 2 ) ; and −1

(8.96)

∆Ω = 0.25π Ω 0 ∆ a ; ( 2 ) −2

p0′ is the target return signal power in the absence of frequency modulation. The parameter dh , as before, deﬁnes the number of radar range zones. In the case of stationary radar and vertical scanning, the power spectral density takes the following form:20 ∞

S(ω) =

X −n p0′ − π dh2 ⋅e {1 − Φ [ (1+ π dπh )(dXh h − n) ]} ⋅ δ[ω − (ω im − nω M )] ⋅ e 0.5 d , | dh| n= − ∞

∑

h

h

(8.97) where Φ(x) is the error integral given by Equation (1.27). Reference to Equation (8.97) shows that if the condition dh << 1 is satisﬁed, the argument in the function Φ(x) is high in value and we can use the following approximation:

1 − Φ(x) =

e−x . x π

(8.98)

Thus, based on Equation (8.97) the power spectral density can be written in the following form:

S(ω ) = p0′ ⋅

∞

∑ δ[ω − (ω

− nω M )] ⋅ e − π( Xh −n) . 2

im

(8.99)

n=−∞

The formula in Equation (8.99) coincides with Equation (8.70). In other words, under the condition dh << 1, the power spectral density is independent of the characteristics of the directional diagram and is only deﬁned by the modulating function. Reference to Equation (8.97) shows that with an increase in the value of dh , i.e., in the number of radar range zones, the power spectral density becomes asymmetric with an increase in n relative to n0. In the opposite direction, the power spectral density decreases according to the function sinc2 π(Xh – n) in Equation (8.94), which can be explained in the following manner. The radar range ρ can only increase with respect to the altitude h within the directional diagram. Consider some examples of the power spectral densities under the condition dh > 1 in the case of the symmetric sawtooth law of frequency modulation. Consider the case of moving radar. Reference to Equation (8.89) and EquaΩ tion (8.94) shows that if the conditions dh << 1 and|∆Ω| << |Xh0 |are satisﬁed,

Copyright 2005 by CRC Press

1598_book.fm Page 342 Monday, October 11, 2004 1:57 PM

342

Signal and Image Processing in Navigational Systems

we can neglect the second term in the function sinc2 x, and the power spectral density of the Doppler ﬂuctuations at the frequency Ωn takes place. Under 0 the condition dh >> 1, the bandwidth of the partial power spectral density of the ﬂuctuations is deﬁned by the function

sinc

2 π ( Ωn − ω ) Xh Ω0

[

]= e

−

π ( Ωn − ω )2 Xh2 Ω02

=e

−

π ( Ωn − ω )2 dh2 2 ∆Ω2

.

(8.100)

The bandwidth of this function is much less than the bandwidth of the power spectral density of the Doppler ﬂuctuations. Because the power spectral density of the Doppler ﬂuctuations is one-sided under the condition Xh = n0 , the product of the function g˜ 2 (ψ ) and Equation (8.100) deﬁnes the onesided power spectral density of the ﬂuctuations. Under the condition dh >> 1, the maximums of other partial power spectral densities of the ﬂuctuations take place at the frequencies

ω k = Ω n0 + k − k ⋅

Ω0 Ω0 = Ω n0 − kω M +1 Xh Xhω M

[

]

(8.101)

when the function sinc2 x = 1. The frequency ωk , as with sloping scanning, differs from the frequency Ωn +k. This can be explained, as before, by the fact 0 that each n-th harmonic is formed by an area (in the present case by a ring) corresponding to the resolved area of radar with the frequency-modulated searching signal. In the present case, the Doppler isofrequency lines are circles (see Figure 8.2). The k-th ring resolved with respect to the circle has a kΩ Doppler shift equal to Xh0 at n0 , which is subtracted from the frequency Ωn +k 0 in Equation (8.101). Substituting Equation (8.101) in Equation (8.89) or Equation (8.94), we can write the power spectral density of the ﬂuctuations in the following form: ∞

∑ (

2π g 2 |Ω0 |n = − ∞

S(ω) = p0 ⋅

Ω ωk + k 0 – ω Xh

0.5 Ω0

)⋅ siinc [(ω − ω) 2

k

Xh Ω0

].

(8.102)

As noted above, the partial power spectral density at the frequency Ωn (k = 0) 0 is one-sided and its shape is deﬁned by the function sinc2 x in Equation (8.102) under the condition dh >> 1, i.e., the shape is the one-sided Gaussian curve. Under the condition k ≠ 0, the functions g˜ 2 (ϕ, ψ) and sinc2 x are shifted kΩ in argument relative to the frequency Ωn +k by the value Xh0 . At the point ω 0 2 2 = ωk , the functions g˜ (ϕ, ψ) and sinc x are not one-sided and the partial power spectral density takes the shape of the two-sided Gaussian curve with average frequency ωk . Substituting Equation (8.101) in Equation (8.94), we can deﬁne the envelope of partial power spectral densities in the following form: Copyright 2005 by CRC Press

1598_book.fm Page 343 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

2 π ˜2 S(ω k ) = p0 ⋅ g |Ω 0 |

( ). 2k Xh

343

(8.103)

Thus, the value of Xkh is always positive due to the one-sided power spectral density of the Doppler ﬂuctuations. When Xh is negative, k must be negative, too, and only frequencies (n0 – k)wM are in the power spectral density. If the condition Xh > 0 is satisﬁed, we can observe the same picture, but at frequencies (n0 + k)wM. The difference in average frequencies of two neighboring partial power spectral densities is determined by

ω k − ω k +1 = ω M ( ω MΩ0Xh + 1) .

(8.104)

The condition of compression of all partial power spectral densities has the following form:

Ω0 = − ω M Xh = − n0ω M ,

(8.105)

i.e., the frequencies Ω0 and nωM must have the same sign as the frequency Ωn . The compressed power spectral density is the one-sided power spectral density at the frequency Ωn and the sum of two-sided power spectral densities at the same frequency. In the absence of frequency modulation under the condition Ω0 > 0, i.e., as the radar moves closer to the Earth’s surface, the maximum of the power spectral density of the Doppler ﬂuctuations is at the frequency ωim + Ω0 (see Figure 8.13b). The power spectral density of the ﬂuctuations with the frequency-modulated searching signal is shown in Figure 8.13a under the condition Xh > 0, i.e.,

ω k = ω im + Ω 0 − (n0 + k )ω M .

(8.106)

The bandwidth of the partial power spectral densities is equal to 2∆Ω dh . The compressed power spectral density is shown in Figure 8.14 on a decreased scale, under the condition Xh < 0. This power spectral density is obtained by superposition of all partial power spectral densities in the frequency region

ω im + 2Ω 0 = ω im + 2n0 ω M .

(8.107)

It should be noted that under vertical scanning and with ε0 ± 90°, the values of dh and ∆Ω are much less in comparison with the case of sloping scanning if the velocity vector of moving radar is outside the directional diagram. For instance, at ∆fM = 20 MHz, ∆ = 0.1 (6°), and h = 750 m, we get Xh = 100 m and dh = 0.15. In the present case, the power spectral density Copyright 2005 by CRC Press

1598_book.fm Page 344 Monday, October 11, 2004 1:57 PM

344

Signal and Image Processing in Navigational Systems

S (ω) Smax (ω) 1.00

0.75 (a)

(b)

0.50

0.25

ω4

ω3

ω2

0.00 ω1 ωim + Ω0 −nωM 0.0

ω−Ω ∆Ω 0.5

1.0

1.5

2.0

ωim + Ω0

FIGURE 8.13 Power spectral densities of target return signal ﬂuctuations with vertically moving radar and vertical scanning: (a) the asymmetric saw-tooth frequency-modulated searching signal; (b) frequency modulation is absent: dρ >> 1.

1.00

S (ω) S max (ω)

0.50

ω ω0 + 2 Ω0 FIGURE 8.14 Power spectral densities of target return signal ﬂuctuations with vertically moving radar and vertical scanning, with the asymmetric saw-tooth frequency-modulated searching signal. Condition of compression of the power spectral density is satisﬁed.

corresponds to the condition dh << 1 and consists only of a single power spectral density of the Doppler ﬂuctuations at the frequency Ωn0 ; deﬁnition of the average frequency of the target return signal is not difﬁcult. Consider now another case that is very important in practice.

Copyright 2005 by CRC Press

1598_book.fm Page 345 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target 8.3.3

345

Vertical Scanning: The Velocity Vector Is Outside the Directional Diagram

Assume that the directional diagram is directed vertically down. The velocity vector of moving radar is close to the horizontal direction. Suppose that the axial symmetric directional diagram and the speciﬁc scattering area S° are deﬁned by Gaussian surfaces. The power spectral density of the ﬂuctuations can be written in the following form: ∞

S(ω) =

p0′ π|dh ∆Ω|n = − ∞

×e

∑

[

− π Xh − n + πdh

Ωn ) 1 + πdh[Xh − n + πdh ⋅ (ω2−∆Ω 2 ]⋅e 2

−π

( ω − Ωn )2 ∆Ω2

, 2

( ω − Ωn ) 2 ∆Ω2

]

K0.25 (z 2 ) ⋅ e z

2

(8.108) where

z = 2

2

{1 + πdh[Xh − n + πdh ⋅ (ω2−∆ΩΩ ) ]} 2 n 2

2πdh2

;

(8.109)

p0′ is the target return signal power in the absence of frequency modulation [see Equation (2.203)];

Ω 0 = 4 πVλ−1 sin ε 0 ;

(8.110)

∆Ω = 2 2 πV λ −1 a−2 cos ε 0 ;

(8.111)

K0.25 (x) = ( 2 )−1 π ⋅[I− 0.25 (x) − I0.25 (x)]

(8.112)

is the modiﬁed Bessel function of the second kind; and I±0.25 (x) is the modiﬁed Bessel function of the ﬁrst kind. The functions I±0.25 (x) are tabulated. Under the condition x < 0.1, we can use a very close approximation for the functions I±0.25 (x). Thus, we can write K 0.25 ( x) = 2.158 ⋅ (x −0.25 − 0.9596 ⋅ x 0.25 ) .

(8.113)

If the required accuracy is not more than 15%, we can use an approximation given by Equation (8.113) under the condition x ≤ 0.5. With high values of x, we can use the following approximation: Copyright 2005 by CRC Press

1598_book.fm Page 346 Monday, October 11, 2004 1:57 PM

346

Signal and Image Processing in Navigational Systems

K 0.25 (x) = 0.5πx −1 ⋅ e − x .

(8.114)

Reference to Equation (8.108) shows that the maximum of the partial power spectral density under the conditions

ω = Ω n0

X h = n0

and

(8.115)

has the following form:

S(Ω n0 ) = At the frequency Ωn

0+k

1 p0′ ⋅ K 0.25 ( 2 π1d2 ) ⋅ e 2 π d . h π|dh ∆Ω| 2 h

(8.116)

, we can write

2 2 p0′ S(Ωn0 + k ) = ⋅ K [ (1 − π dhk ) ] ⋅ |1 − πdh k|⋅ e − π k ⋅ e π|dh ∆Ω| 0.25 2 π dh2

( 1 − π dh k )2 2 π dh2

. (8.117)

The maximums of the partial power spectral densities are at the frequencies given by the condition

Xh − n + π ⋅

dh (ω − Ω n )2 =0. 2∆Ω 2

(8.118)

Reference to Equation (8.118) shows that there are two maximums at the frequencies equal to

ω ′k = Ω n0 + k + 2π −1kdh−1

ω ′′k = Ω n0 + k − 2π −1kdh−1

and

(8.119) for each partial power spectral density. As one can see from Equation (8.119), the frequencies ωk′ and ωk″ are symmetric with respect to the frequency Ωn0+k . The parameters k and dh in Equation (8.119) must have the same sign because the power spectral density, as in Section 8.3.1, is one-sided. Substituting Equation (8.119) in Equation (8.103), we can deﬁne these maximums

S(ω k ) =

1 − 2k − 2k p0′ ⋅ K 0.25 ( 2 π1d2 ) ⋅ e d ⋅ e 2 π d = S(Ω n0 ) ⋅ e d . h π|dh ∆Ω| h

2 h

h

(8.120)

The partial power spectral densities are shown in Figure 8.15–Figure 8.19 for various values of k and dh . At x ≤ 0.1, the approximation given by Equation (8.113) is used. Copyright 2005 by CRC Press

1598_book.fm Page 347 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

347

S (ω) p ∆Ω 1.0 1 2 3 0.5 4

5 x0 0

0.5

1.0

FIGURE 8.15 Partial power spectral densities of target return signal ﬂuctuations with the frequency-modulated searching signal and vertical scanning. The velocity vector of moving radar is outside the ω − ( ω + Ω′ − n ω ) radar antenna directional diagram, k = 0 , x 0 = : (1) dh = 0.1; (2) dh = 0.3; (3) dh = 1; ∆Ω (4) dh = 3; (5) dh = 10. im

0

0

M

S (ω) p ∆Ω 0.10

0.05 1 2 3 4 x−1 0

0.5

1.0

FIGURE 8.16 Partial power spectral densities of target return signal ﬂuctuations with the frequency-modulated searching signal and vertical scanning. The velocity vector of moving radar is outside the ω − [ ω + Ω′ − ( n − 1 ) ω ] radar antenna directional diagram, k = −1, x −1 = : (1) dh = 0.1; (2) dh = 0.3; (3) dh ∆Ω = 1; (4) dh = 3. im

Copyright 2005 by CRC Press

0

0

M

1598_book.fm Page 348 Monday, October 11, 2004 1:57 PM

348

Signal and Image Processing in Navigational Systems

S(ω) p ∆Ω 0.3

0.2 3 4

5

0.1

2 1 x1 0

0.25

0.50

0.75

FIGURE 8.17 Partial power spectral densities of target return signal ﬂuctuations with the frequency-modulated searching signal and vertical scanning. The velocity vector of moving radar is outside the ω − [ ω + Ω′ − ( n + 1 ) ω ] radar antenna directional diagram, k = 1, x1 = : (1) dh = 0.1; (2) dh = 0.3; (3) dh = ∆Ω 1; (4) dh = 3; (5) dh = 10. im

0

0

M

S(ω) p ∆Ω 0.2 2

3

0.1

1 x2 0.20 0.25

0.50

0.75

1.00

FIGURE 8.18 Partial power spectral densities of target return signal ﬂuctuations with the frequency-modulated searching signal and vertical scanning. The velocity vector of moving radar is outside the ω − [ ω + Ω′ − ( n + 2 ) ω ] radar antenna directional diagram, k = 2 , x 2 = : (1) dh = 1; (2) dh = 3; (3) dh = 10. ∆Ω im

Copyright 2005 by CRC Press

0

0

M

1598_book.fm Page 349 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

349

S(ω) p ∆Ω 0.2

2

0.1

1

x3 0.25

0.50

0.75

1.00

FIGURE 8.19 Partial power spectral densities of target return signal ﬂuctuations with the frequency-modulated searching signal and vertical scanning. The velocity vector of moving radar is outside the ω − [ ω + Ω′ − ( n + 3 ) ω ] radar antenna directional diagram, k = 3 , x 3 = : (1) dh = 3; (2) dh = 10. ∆Ω im

0

0

M

Only positive branches of the power spectral densities are presented in Figure 8.15–Figure 8.19 at xk > 0. At dh > 1, the shape of the one-half partial power spectral densities is not symmetric: in the case ω < ωk , they are more slanting in comparison with the case ω > ωk. With an increase in the value of dh, this asymmetry decreases. At dh = 0.1 or 0.3, the power spectral density bandwidth approximately coincides with the power spectral density bandwidth of the Doppler ﬂuctuations under the condition k = 0, which is deﬁned by the Gaussian law. At high values of dh , the power spectral density bandwidth is deﬁned by

1 |dh|

. At k ≥ 1 and dh ≥ 1, the bandwidth of the power

spectral densities is determined by

∆ω k =

∆Ω . 2πkdh

(8.121)

The power spectral densities under the condition ∆Ω = ωM are shown in Figure 8.20 and Figure 8.21. At dh = 0.3, the partial power spectral densities under the conditions k = 0 and k = 1 are symmetric and are at the frequency Ωn +k . At dh = 3, each partial power spectral density is divided into two parts 0 in accordance with Equation (8.119). Neighboring partial power spectral densities partially overlap. The normalized envelope S(ωk) given by Equation (8.120) is shown in Figure 8.21 by the dotted line.

Copyright 2005 by CRC Press

1598_book.fm Page 350 Monday, October 11, 2004 1:57 PM

350

Signal and Image Processing in Navigational Systems

1.0

S(ω) ∆Ω p

0.5

x n0 + 1

n0 − 1

n0

FIGURE 8.20 Power spectral densities of target return signal ﬂuctuations with the asymmetric saw-tooth frequency-modulated searching signal and vertical scanning. The velocity vector of moving radar is outside the radar antenna directional diagram: ωM = ∆Ω and dh = 0.3.

1.0

S(ω) ∆Ω p

0.5 Sk (ω) ∆Ω p

x n0 + 3

n0 + 2

n0 + 1

n0

n0 − 1

FIGURE 8.21 Power spectral densities of target return signal ﬂuctuations with the asymmetric saw-tooth frequency-modulated searching signal and vertical scanning. The velocity vector of moving radar is outside the radar antenna directional diagram: ωM = ∆Ω and dh = 3.

8.4

The Symmetric Saw-Tooth Frequency-Modulated Searching Signal

In the case of symmetric saw-tooth frequency modulation, the coherent searching signal can be determined by Equation (8.5), where

Copyright 2005 by CRC Press

1598_book.fm Page 351 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

351

 Ψ(t − nTM ) = 0.5kω (t − nTM )2 at nTM < t < (n + 0.5)TM ;    Ψ[t − (n + 1)TM ] = − 0.5kω [t − (n + 1)TM ]2 at (n + 0.5)TM < t < (n + 1)TM . (8.122) We have, respectively (see Figure 8.22),

ω 1 (t − nTM ) = ω 0 + kω (t − nTM ) and ω 2 [t − (n + 1)TM ] = ω 0 − kω [t − (n + 1)TM ]

.

(8.123)

With the same frequency deviation ∆ωM and the same period TM, the parameter kω , in the present case, is twice that in the case of asymmetric saw-tooth frequency modulation:

kω = 2∆ω M T −1 .

(8.124)

M

To estimate the correlation function of the ﬂuctuations it is necessary, ﬁrst, to deﬁne the transformed elementary signals given by Equation (8.8) and Equation (8.9) within the limits of the various intervals t1, …, t5 (see Figure 8.23) belonging to the period TM; second, to deﬁne the differences ∆Ψ given by Equation (8.35); third, to use the Fourier-series expansion for the function ej∆Ψ in accordance with Equation (8.37); fourth, to substitute results from the Fourier-series expansion into the correlation function of the ﬂuctuations, and to carry out averaging over the period TM as in Section 8.3. As one can see from Figure 8.24, the frequency of the transformed target return signal within the limits of the intervals [t2, t3] and [t4, t5] varies linearly in time. To keep the sign of the frequency shift kωTd constant, the parameter ωim is introduced in Figure 8.24. In principle, we can deﬁne the correlation function of the ﬂuctuations taking these intervals into consideration. However, if the condition Td << TM is true, we can neglect these intervals (as in Section 8.3), as the power they contribute is proportional to

4 Td2 2 TM

. In the case of radar with

the frequency-modulated searching signal, the receiver does not operate at this instant of time and the target return signals are not processed. Using the same mathematics as in Section 8.3, we can write the correlation function of the ﬂuctuations in the following form:

R(τ) = 0.25p0′ ⋅ e jωimτ

∫∫

g 2 (ϕ, ψ ) ⋅ e

−2 j ( ω 0 + 0.5 ∆ω M )

∆ρ c

∞

×

∑ [sinc π(X 2

ρ

− 0.5n) + sinc 2 π(Xρ + 0.5n)] ⋅ e − jnω τ dϕ dψ, M

n= − ∞

(8.125) Copyright 2005 by CRC Press

1598_book.fm Page 352 Monday, October 11, 2004 1:57 PM

352

Signal and Image Processing in Navigational Systems ω

ω1(t )

ω 2(t )

ω0 −TM

t 0

TM

FIGURE 8.22 The frequency of the searching signal under symmetric saw-tooth frequency modulation. ω

1

2

t2

t3

ω0 0

t0

t1

t4

t5

t

FIGURE 8.23 The frequency of the heterodyne (1) and target return (2) signals under symmetric saw-tooth frequency modulation.

where p0′ and Xρ are the same as in Equation (2.203) and Equation (3.12), but the value of Xρ is positive. Unlike in Equation (8.52), we have two sinc2 x functions. The function sinc2 (Xρ – 0.5n) is different from zero at positive values of n; the function sinc2 (Xρ + 0.5n) is different from zero at negative values of n. Furthermore, unlike in Equation (8.52), the function sinc2 x has as an argument 0.5n instead of n. This is because the value of kω is twice that in the case of asymmetric saw-tooth frequency modulation. The range-ﬁnder frequency at the radar range ρ is double too: Copyright 2005 by CRC Press

1598_book.fm Page 353 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

353

ωh − ωr TM

kωTd ωim

t1

t2

t3

t4

t5

t

kωTd

FIGURE 8.24 The frequency of the transformed target return signal under symmetric saw-tooth frequency modulation.

Ωρ = 2Xρω M = n0ω M .

(8.126)

The functions sinc2 x in Equation (8.125) are twice as wide as those in Equation (8.52). The bandwidth of the functions sinc2 x in Equation (8.125) is equal to two in normalized units. Taking the previously mentioned peculiarities into consideration, based on Equation (8.52) and Equation (8.125) the power spectral density of the ﬂuctuations takes the following form:

S(ω ) = 0.25[Sh (ω − Ω ′n , Xρ0 − 0.5n) + Sh (ω − Ω ′′n , Xρ0 + 0.5n)] , (8.127) where

Ω ′n = ω im + Ω ′0 − nω M

Ω ′′n = ω im + Ω ′0 + nω M . (8.128)

and

Using Equation (8.127), we can extend all the results discussed in Section 8.3 to the case of symmetric saw-tooth frequency modulation. Let us consider the main characteristics of the power spectral density of the ﬂuctuations under sloping scanning. In the case of stationary radar and with the use of the approximation given by Equation (8.60), the power spectral density of the ﬂuctuations takes the following form: ∞

S(ω) = 0.25pFM

∑ δ[ω − (ω

im

− nω M )] ⋅ e

n= − ∞

Copyright 2005 by CRC Press

( Xρ′ − 0.5 n )2 0

1 + dρ2

. 2

+ δ[ω − (ω im + nω M )] ⋅ e

−π

−π

( Xρ′ + 0.5 n ) 0

1+ dρ2

(8.129)

1598_book.fm Page 354 Monday, October 11, 2004 1:57 PM

354

Signal and Image Processing in Navigational Systems

Under the condition dρ << 1, where dρ is given by Equation (8.56), the power spectral density of the ﬂuctuations has six discrete components (see Figure 8.25); other discrete components are inﬁnitesimal. If the condition 2Xρ = n0 is true, the frequencies of these discrete components are determined 0

by Equation (8.128), i.e., at Ω0 = 0, we have n0 – 1, n0 , and n0 + 1. The relative power of each discrete component at the frequency ωim + n0ωM is equal to one. The relative power of other discrete components is equal to 0.456 or less. The total power of all discrete components is equal to 0.956, and not 1, as we take into consideration only terms included in the main lobe of the function sinc2 x.The power spectral density is shown in Figure 8.25 for the − π ( X ′ ∓ 0.5 n)2

ρ0 given case. The envelope e is shown by the dotted line. The condition dρ << 1 corresponds, as before, to the condition δρ0 >> δρ (see Figure 8.7), as radar resolution with a frequency-modulated searching signal depends only on the frequency deviation ∆fM. Under the condition dρ >> 1, the power spectral densities correspond to the condition δρ0 << δρ (see Figure 8.26). Because the target return signals from each radar range zone are noncoherent, all side components in Figure 8.25 are summarized power. The effective power spectral density bandwidth with the range-ﬁnder frequency is equal to 2dρωM. Each radar range zone corresponds to two harmonics. Because the total bandwidth is equal to 2dρωM , the number of radar range zones is equal to dρ , as in the case of asymmetric saw-tooth frequency modulation.

S (ω) pFM 0.25

ω n0 − 1 n0 + 1 ωim − n0 ωM

ωim

n0 − 1

n0 + 1

ωim + n0 ωM

FIGURE 8.25 The power spectral density of target return signal ﬂuctuations with the symmetric saw-tooth frequency-modulated searching signal: dρ << 1.

Copyright 2005 by CRC Press

1598_book.fm Page 355 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

355

S (ω) pFM 0.25

ω ωim

ωim − n0 ωM

ωim + n0 ωM

FIGURE 8.26 The power spectral density of target return signal ﬂuctuations with the symmetric saw-tooth frequency-modulated searching signal: dρ >> 1.

Let us deﬁne the power spectral density of the ﬂuctuations with moving radar. We should take into consideration that the envelope S(ωk) for an arbitrary shape of the function g˜ 2 (ϕ, ψ) is determined by Equation (8.78), replacing dρ with 2dρ. The formula in Equation (8.61) for the case of symmetrical saw-tooth frequency modulation has the following form ∞

∑ [e

pFM S(ω) = 4|∆ΩvFM |n = − ∞

−π

( ω − Ω′n )2 ∆Ω2

v FM

−π

+e

( ω − Ωn′′ )2 ∆Ω2

v FM

]⋅ e

−π

( Xρ′ − 0.5|n|)2 0

1 + dρ2

, (8.130)

where the parameters pFM , ∆ΩvFM , dρ , and Xρ′0 are the same as in Equation (8.61), but the values of dρ and Xρ′ are always positive. The values of Ωn′ and 0 Ωn″ are determined by

Ωn = ω im + Ω′0 −

∆Ωv dρXρ0 1 + dρ2

[

+n

∆Ωv dρ 2 (1 + dρ2 )

]

∓ ωM ,

(8.131)

where the sign “–” at the frequency ωM corresponds to Ωn′ , i.e., to the frequency ωim + Ω0′ – nωM; the sign “+” corresponds to the frequency ωim + Ω0′ + nωM. Reference to Equation (8.131) shows that the frequency Ωn′ is independent of n if the condition

ωM =

Copyright 2005 by CRC Press

∆Ω v dρ 2(1 + dρ2 )

(8.132)

1598_book.fm Page 356 Monday, October 11, 2004 1:57 PM

356

Signal and Image Processing in Navigational Systems

is true. In the present case, the partial power spectral densities of the ﬂuctuations are superimposed on one another, i.e., the power spectral density is compressed. In accordance with Equation (8.132), this effect takes place under the condition ∆Ωv > 0 or α = β0 = 0° if the equality ∆Ωv = 2dρωM is true, because dρ2 ≈ 1 + dρ2 . The average frequency of the compressed power spectral density is determined by

Ω ′n = ω im + Ω ′0 − 2ω M Xρ0 = ω im + Ω ′0 − n0ω M .

(8.133)

Power spectral density compression is absent in the total frequency Ω ′′n and the difference between the average frequencies of neighboring partial power spectral densities is equal to 2ωM. If the directional diagram is directed back, in accordance with Equation (8.58) the frequency Ω ′0 , as well as Ωv and ∆Ωv , would be negative. In the present case, power spectral density compression is possible within the limits of the other one-half modulation period, as ωM and ∆Ωv in Equation (8.131) have different signs. With a further increase in the value of dρ , the number of partial power spectral densities increases and the difference of their average frequencies tends to approach the frequency ωM. The effective bandwidth of the compressed power spectral density is equal to 2ωM, i.e., it corresponds to the effective power spectral density bandwidth in the case of stationary radar. It is not difﬁcult to deﬁne the power spectral density S(ω) under the condition β0 ≠ 0°.21,22 For this purpose, it is necessary to use Equation (8.61) and Equation (8.127), and to replace ∆ΩFM with the value of ∆ΩFM given by Equation (8.67). If the condition of power spectral density compression of the ﬂuctuations is satisﬁed, the bandwidth ∆ΩFM is determined by

∆Ω FM = 4ω 2M + ∆Ω 2h .

(8.134)

The partial power spectral densities are shown in Figure 8.27. The total power spectral density is shown in Figure 8.28. The power spectral densities at the total and difference frequencies have approximately the same bandwidth, which is close to ∆Ω. In an analogous way, we can consider the power spectral density under vertical scanning and vertically moving radar [see Equation (8.89) and Equation (8.90)]. The main peculiarity of this power spectral density is that the partial power spectral densities are asymmetrical at n0 – 1 and n0 + 1. The power spectral densities for the cases of stationary radar and vertical scanning in the presence of the frequency-modulated searching signal are shown in Figure 8.29–Figure 8.31 for various values of dh. The power spectral densities are one-sided in the direction of increasing n. The left part of the power spectral density corresponds to a decrease in the function sinc2 x. Figure 8.29 corresponds to reﬂection from one radar range zone, and Figure

Copyright 2005 by CRC Press

1598_book.fm Page 357 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

357

S (ω) Smax(ω) 1.00

0.75

0.50

0.25 n 29.5

37.5

45.5

FIGURE 8.27 Partial power spectral densities of target return signal ﬂuctuations with the symmetric sawtooth frequency-modulated searching signal at total and difference frequencies: Xρ ωM < Ω0; dρ 0 < 1; Ω0 = 37.5 ωM; Xρ = 4; ∆v = ∆h = 0.1; γ0 = 65°; and β0 = 45°. 0

S Σ (ω) 2.00

1.75 1.50

1.25

1.00

0.75

0.50

0.25 n 29.5

37.5

45.5

FIGURE 8.28 The total power spectral density of target return signal ﬂuctuations with the symmetric sawtooth frequency-modulated searching signal at total and difference frequencies: Xρ ωM < Ω0; dρ 0 < 1; Ω0 = 37.5 ωM; Xρ = 4; ∆v = ∆h = 0.1; γ0 = 65°; and β0 = 45°. 0

Copyright 2005 by CRC Press

1598_book.fm Page 358 Monday, October 11, 2004 1:57 PM

358

Signal and Image Processing in Navigational Systems S (ω) Smax(ω) 1.00 0.75 0.50 0.25 ωim

n 78

79

80

81

82

FIGURE 8.29 The power spectral density of target return signal ﬂuctuations with the symmetric saw-tooth frequency-modulated searching signal and stationary radar: ∆fM = 20 MHz; ∆v = 0.1; h = 300 m; Xh = 40.

S (ω) Smax(ω) 1.00 0.75 0.50 0.25 ωim

n 798

799

800

801

802

803

804

FIGURE 8.30 The power spectral density of target return signal ﬂuctuations with the symmetric saw-tooth frequency-modulated searching signal and stationary radar: ∆fM = 20 MHz; ∆v = 0.1; h = 3 km; Xh = 400.

8.30 corresponds to reﬂection from two radar range zones. Figure 8.31 corresponds to reﬂection from four radar range zones.

8.5

The Harmonic Frequency-Modulated Searching Signal

In the case of the harmonic frequency-modulated searching signal, we can write

Copyright 2005 by CRC Press

1598_book.fm Page 359 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

359

S (ω) S max (ω) 1.00 0.75 0.50 0.25 n

ωim

3999

4000

4001

4002

4003

4004

4005

FIGURE 8.31 The power spectral density of target return signal ﬂuctuations with the symmetric saw-tooth frequency-modulated searching signal and stationary radar: ∆fM = 20 MHz; ∆v = 0.1; h = 15 km; Xh = 2000.

Ψ(t) =

∆ω M ⋅ cos ω M t . ωM

(8.135)

The power spectral density of the nontransformed ﬂuctuations (averaged over time) with harmonic frequency modulation is determined by Equation (2.70), where |N n |2 = J n2

(

∆ω M ωM

).

(8.136)

The correlation function of the nontransformed ﬂuctuations (averaged over time) is given by Equation (8.14) where Cρn− 0.5 ∆ρ = J n[

2 ∆ω M ωM

⋅ sin

ω M ( ρ − 0.5 ∆ρ) c

]

and Cρn+ 0.5 ∆ρ = J n[ 2 ω∆ωMM ⋅ sin ω M (ρ +c 0.5 ∆ρ) ] . (8.137)

Neglecting the Doppler shift in frequency ωM , based on Equation (8.14) and in terms of Equation (8.135), the correlation function can be written in the following form ∞

R(τ) = p0

∑e

j ( ω im + nω M ) τ

n= − ∞

∫∫

g 2 (ϕ , ψ ) J n2 ( Mρ ) ⋅ e

−2 jω 0

∆ρ c

dϕ dψ ,

(8.138)

where Mρ =

Copyright 2005 by CRC Press

2∆ω M ω ρ ⋅ sin M . c ωM

(8.139)

1598_book.fm Page 360 Monday, October 11, 2004 1:57 PM

360

Signal and Image Processing in Navigational Systems

Consider some speciﬁc cases under sloping scanning. Assume that the radar moves horizontally, i.e., ε0 = 0. Suppose that the condition ωM δρ << 1 T [the condition TMr << 1 and δρ are determined by Equation (8.71)] is satisﬁed. This case corresponds to a point target. Based on Equation (8.138) we can write ∞

Sh (ω ) =

∑ J (M 2 n

n=−∞

ρ0

)δ[ω − (ω im + nω M )] ,

(8.140)

where S(ω) is the power spectral density in the absence of frequency modulation, with ω0 replaced by ωim + nωM. We use Equation (8.140) to deﬁne allowable spurious harmonic frequency modulation caused, for example, by low-frequency pulsed power supply voltage in the super high-frequency radar generator with continuous nonmodulated searching signals. Consider two cases: ∆Ω > ωM and ∆Ω < ωM. In the ﬁrst case, with an increase in Mρ , the power spectral density initially 0

expands; its maximum decreases and after that it becomes two-mode and further, multimode, but it remains continuous (see Figure 8.32). The com∆Ω

ω

puter-modeling ratio ∆ΩFM as a function of ∆ΩM with one-mode power spectral density and with an allowable extension of 50% is shown in Figure 8.33. For instance, assume that h = 15 km, γ0 = 65°, F0 = 10 kHz, ∆F = 1.5 kHz, and fM = 400 Hz. At

ωM ∆Ω

= 0.27, based on Figure 8.33 we can ﬁnd that Mρ = 2 and 0

the allowable value of frequency deviation ∆fM is equal to 3.0 kHz. In the case ∆Ω < ωM , the power spectral densities S[ω – (ωim + nωM)] are not close to one another. Given the ratio

J 02 J n2

, we can ﬁnd the allowable frequency

deviation. For instance, at h = 3 km, γ0 = 65°, fM = 400 Hz, and

J 02 J n2

= 30, the

allowable frequency deviation is equal to 5.3 kHz. ω ρ Assume that Mc << 1 and 2πXρ0 << 1, i.e., that there is a low radar range or low frequency modulation at a not so high frequency deviation. Then, J n2 [

2 ∆ω M ωM

⋅ sin

ω Mρ c

]≈

(2 πXρ0 )2|n| 2 2|n|(|n|!)2

⋅ (1 + 2|n|ψ ctg γ 0 ) .

(8.141)

Substituting Equation (8.141) in Equation (8.138), we can deﬁne the correlation function and power spectral density. With the Gaussian directional diagram, the power spectral density takes the following form:23 ∞

∑

p0′ − π S(ω ) = ⋅e ∆Ω n= − ∞

Copyright 2005 by CRC Press

(ω - Ω′n )2 ∆Ω2

,

(8.142)

1598_book.fm Page 361 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

361

S (f ) 1.00

1

2

3 0.25

f, kHz 1.5

2.0

2.5

3.5

3.0

4.0

4.5

FIGURE 8.32 Power spectral densities of target return signal ﬂuctuations with the harmonic frequencymodulated searching signal, fM = 400 kHz: (1) ∆fM = 0 kHz; (2) ∆fM = 4 kHz; (3) ∆fM = 8 kHz.

∆Ω FM ∆Ω

4

2

3

2.00 1.75 1

1.50 1.25

ωM ∆Ω

1.00 0.0

0.5

1.0

FIGURE 8.33 ∆Ω Relative expansion ∆ΩFM of power spectral density of target return signal Doppler ﬂuctuations with the frequency-modulated searching signal and various values of Mρ ; ∆Ω is the power 0 spectral density bandwidth of the target return signal Doppler ﬂuctuations in the absence of frequency modulation : (1) Mρ = 1; (2) Mρ = 1.5; (3) Mρ = 2; (4) Mρ = 5. 0

Copyright 2005 by CRC Press

0

0

0

1598_book.fm Page 362 Monday, October 11, 2004 1:57 PM

362

Signal and Image Processing in Navigational Systems

where p0′ n =

p0′ (2 πXρ0 )2|n| 2 2|n|(|n|!)2

;

(8.143)

Ω ′0 = Ω 0 (1 − δ 1 − δ 2 + δ n ) ;

(8.144)

δ n = 0.5π −1 |n|∆(v2 ) ;

(8.145)

p0′ is the target return signal power in the absence of frequency modulation; δ1 is the shift due to the speciﬁc effective scattering area S° in the absence of frequency modulation; δ2 is the shift due to radar range in the absence of frequency modulation; δn is the shift caused by frequency modulation. Let us assume that

ω Mρ c

< 1 and 2πXρ0 >> 1, i.e., that there is slow frequency

modulation with high frequency deviation. Under these conditions, we can use the following approximation: − 1 J (2 πXρ ) = 2 ⋅ e π Xρ 2 n

π2 [

α12

2 Xρ − 0.25( 2 n + 1)]

,

(8.146)

where −0.5 ≤ 2Xρ − 0.25(2n + 1) ≤ 0.5 ,

(8.147)

and α1 = 0.78. Using Equation (8.146) and Equation (8.147), the power spectral density with the Gaussian directional diagram can be determined as follows: ∞

∑

pFM − π S(ω ) = ⋅e ∆Ω n= − ∞

(ω − Ω′n )2 ∆Ω2

⋅e

−

π2 α12 ( 1 + dρ2 )

[ 2 Xρ′ 0 − 0.25(2n + 1)]2

,

(8.148)

where pFM is given by Equation (8.62); Ω n = ω im + Ω ′0 +

∆Ω v π dρ [2Xρ′ 0 − 0.25(2n + 1)] α 12 (1 + dρ2 )

+ nω M ;

(8.149)

Ω0′ and Xρ′ are the same as in the case of the saw-tooth frequency-modulated 0 searching signal.

Copyright 2005 by CRC Press

1598_book.fm Page 363 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

363

The partial power spectral densities with the harmonic frequency-modulated searching signal, as well as with the symmetric saw-tooth frequencymodulated searching signal, are at the total and difference frequencies Ωn. Thus, there is a narrowing of the power spectral density at the difference frequency and an expansion at the total frequency. We can deﬁne the condition of power spectral density compression as follows: The sum of the terms with index n in Equation (8.149) should be zero. For the case α = β0 = ε0 = 0°, the power spectral density can be deﬁned in an explicit form without additional simpliﬁcations. If the velocity vector V of moving radar is outside the directional diagram and the variables ϕ and ψ are separable in the function g˜ 2 (ϕ, ψ), the power spectral density can be written in the following form: ∞

S(ω ) =

∑

p1 Ω −ω ⋅ g 2 ( Ωn ) J n2 v |Ωv | n = − ∞

{

2 ∆ω M ωM

[

⋅ sin

ω Mh Ω −ω c sin γ 0 + n

(

Ωv

)

]} ,

(8.150)

where the parameters p1, Ωv , and Ωn are determined by Equation (8.58). The formula in Equation (8.150) is analogous to Equation (8.54). The partial power spectral densities are distorted due to the function J n2 ( x), which is the analog of the function sinc2 x in Equation (8.54). For instance, at the frequency ω h

ω = Ωn and under the condition sin c sinMγ 0 = 0 , Jn = 0 (n ≠ 0) and the power 0

spectral density is zero. The target return signal power with “zero” harmonic at the frequency ωim + Ω0′ is maximal. Due to the directional diagram length, the target return signal power can be different from zero at other Doppler frequencies. As one can see from Figure 8.34, the power spectral density of the Doppler ﬂuctuations is highly distorted at the condition β0 = 0° in comparison with the power spectral density under the condition β0 = 45°. If the radar antenna is deﬂected in the direction of moving radar, i.e., if the condition β0 ≠ 0° is satisﬁed, the lines for constant values of ρ and isofrequency Doppler lines on the surface of a two-dimensional target are mismatched, as it takes place under the condition β0 = 0°. In this case, we cannot carry out the integration in Equation (8.138) and it is necessary to use numerical integration techniques. Because the Bessel function depends only on the angle ψ, we can use it together with the function g˜ v2 ( ψ ) in the following way: g˜ 12 ( ψ ) = g˜ v2 ( ψ ) ⋅ J n2 ( ψ ) .

(8.151)

Using the function g˜ 12 ( ψ ) under the condition Tr < TM , we can deﬁne a deterioration of the target return signal due to frequency modulation at the n-th harmonic in comparison with the nonmodulated searching signal. In addition, we can deﬁne the shift ∆γ in the average angle γ0 with the Gaussian directional diagram (n > 0)

Copyright 2005 by CRC Press

1598_book.fm Page 364 Monday, October 11, 2004 1:57 PM

364

Signal and Image Processing in Navigational Systems

1.00

S (ω) Smax(ω) 1

2

0.75 0.50 0.25 0.00 1000

f, Hz 2000

3000

5000

4000

FIGURE 8.34 Experimental power spectral densities of target return signal Doppler ﬂuctuations with the harmonic frequency-modulated searching signal, fM = 1 MHz, ∆fM = 2.4 MHz, n = 3: (1) β0 = 0°; (2) β0 = 45°.

dn ( ThM ) = J n2 ( Mρ ) +

∆γ = −

∆(v2 n) [ J 22 ( Mρ )] ( 2 n) 2 3 n n! π n

∆(v2 ) [ J n2 ( Mρ )] (1) 4 πdn ( ThM )

;

,

(8.152)

(8.153)

where [ J n2 ( Mρ )] ( 2 n) is a derivative of the second order with respect to γ. We need use only the term of the derivative that contains the Bessel function J 02 ( Mρ ) of zero order at the neighboring point J n2 ( Mρ ) = 0 , because other terms are negligible. Under the condition n = 1, we can write  [ J 2 ( M )] ( 2 ) = 0.5 J 2 ( M ) ⋅ M ′ 2 ; 0 ρ ρ ρ  1  2 2 (1)  [ J 1 ( Mρ )] = J 1 ( Mρ )[ J 0 ( Mρ ) − J 2 ( Mρ )] ⋅ Mρ′ ;   M ′ = −2 π ∆ω M ⋅ Td ⋅ ctg γ ctg ω Mρ . ωM TM c 0  ρ

(8.154)

An example of the determination of the value of d1 ( ThM ) is shown in Figure 8.35. The computer-modeled and experimental values of the power Pr of the transformed target return signal, which are normalized to the power Pn of noise, are shown in Figure 8.36. The segment of curve 2 in Figure 8.36 corresponding to low values of h, in which the value of Pr is approximately independent of h in accordance with Equation (8.143)–Equation (8.145), can be seen. The experimental dependence of the ratio plotted in Figure 8.37.

Copyright 2005 by CRC Press

Pr Pn

as a function of h is

1598_book.fm Page 365 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target 135

0

365 270 h ___ TM

−5 −10 −15

1

−20 −25 −30 −35

2

−40 −45 −50 h d1(___ ), dB TM

FIGURE 8.35 Deterioration of the target return signal d1 ( ThM ) with the harmonic frequency-modulated search∆ω ing signal at the ﬁrst harmonic of frequency modulation, γ0 = 65°, ∆v = 5°, ω MM = 1.2: (1) onesided frequency; (2) two-sided convoluted frequencies.

Pr , dB Pn 30

25 2 1

20

2 1

15

10

5

0

h, m 150

300

450

600

750

FIGURE 8.36 P Experimental and computer-modeling ratio Pnr as a function of altitude h, γ0 = 65°, ∆v = 5°, 1.2, n = 1, fM = 300 kHz: (1) one-sided frequency; (2) two-sided convoluted frequencies.

Copyright 2005 by CRC Press

∆ω M ωM

=

1598_book.fm Page 366 Monday, October 11, 2004 1:57 PM

366

Signal and Image Processing in Navigational Systems Pr , dB Pn 30

25

20

15

10

5 h, m

0

100

200

300

400

500

600

700

FIGURE 8.37 P Experimental ratio Pr with the harmonic frequency-modulated searching signal at the third n ∆ω harmonic of frequency modulation (two-sided convoluted): γ0 = 65°, ∆v = 5°, fM = 1 MHz, ω MM = 2.4.

8.6

Phase Characteristics of the Transformed Target Return Signal under Harmonic Frequency Modulation

Let us express the target return signal multiplied by the searching signal in the following form:24 ∞

Sr (t) =

∑ S {J (M ) cos[(ω i

ρ

0

im

+ Ω0i )t + ϕ 0i ]

i=−∞ ∞

+

∞

∑ ∑ J (M n

n= − ∞ i= − ∞

[

ρi

[

T

T

+ (−1)n cos (ω im + Ω0i − nω M )t + π n TMdi + ϕ 0i

Copyright 2005 by CRC Press

]

){cos (ω im + Ω0i + nω M )t − π n TMdi + ϕ 0i , (8.155)

]}}

1598_book.fm Page 367 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

367

where ϕ0i is the phase of an elementary signal distributed randomly within the limits of the interval [0, 2π]. Due to the random character of the phase ϕ0i , the phase of each term in the sum is random, too. At the same time, the phase πn TMd varies within the limits of the narrow region deﬁned by the T

directional diagram under the condition Tr << TM. The average phase πn TMd of the target return signal given by Equation (8.155) deﬁnes some average distance between the radar and the two-dimensional (surface) target. To except the random phase ϕ0i , which does not allow us to measure the phase T

πn TMd , we should take the product of the signals with frequencies ωim + Ω0i T

+ nωM and ωim + Ω0i – nωM, as a rule. For this purpose, the signal Sr(t) must be squared. As a result, we can write ∞

Srsq (t) =

∑ S (t ) J ( M 2 i

2 n

i =−∞

ρi

(

T

)

) cos 2 nω M t − πn TMdi .

(8.156)

Denote ϕ i = 2 πn ⋅

Tdi TM

= 4 πn ⋅

h . cTM sin(ψ + γ 0 )

(8.157)

Based on Equation (8.157) we can write Equation (8.156) in the following form: Srsq (t) = E cos(2nω M + ϕ) = X cos(2nω M t) + Y sin(2nω M t) ,

(8.158)

where ∞

X=

∑ S (t)J (M 2 i

2 n

ρi

) cos ϕ i = E cos ϕ ;

(8.159)

)sin ϕ i = E sin ϕ ;

(8.160)

i=−∞

∞

Y=

∑ S (t)J (M 2 i

2 n

ρi

i=−∞

tg ϕ =

Y . X

The average value ϕav of the phase can be determined by

Copyright 2005 by CRC Press

(8.161)

1598_book.fm Page 368 Monday, October 11, 2004 1:57 PM

368

Signal and Image Processing in Navigational Systems

∞

∑ S (t)J (M 2 i

tg ϕ = tg (ϕ 0 + ∆ϕ i ) =

2 n

ρi

)sin ϕ i

i=−∞

,

∞

∑ S (t)J (M 2 i

2 n

ρi

(8.162)

) cos ϕ i

i=−∞

ϕi = ϕi – ϕ0. Averaging is carried out over an ensemble. Assuming where ∆ϕ that the phase deviation ν = ϕi – ϕ0 is not high in value within the directional diagram, i.e., that the condition γ0 < 90° is satisﬁed, based on Equation (8.162) we can write ϕ av = ϕ 0 + ∆ϕ ,

(8.163)

where ∞

∑ S (t)J (M 2 i

∆ϕ =

2 n

ρi

)∆ϕ i

i=−∞ ∞

.

∑ S (t)J (M 2 i

2 n

ρi

(8.164)

)

i=−∞

Reference to Equation (8.157) shows that ϕ=−

ϕi − ϕ0 ν =− ; ϕ 0 ctg γ 0 ϕ 0 ctg γ 0

(8.165)

4 πnh , cTM sin γ 0

(8.166)

ϕ0 =

where the angle γ0 corresponds to the direction of the directional diagram axis for transmission and reception. In terms of Equation (8.166), we can express Equation (8.164) in the following form: ν2

∆ϕ ϕ=

∫ νA(ν) dν ν1 ν2

∫ A(ν) dν ν1

Copyright 2005 by CRC Press

,

(8.167)

1598_book.fm Page 369 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

369

where A(ν) is the integrand function ψ(ν) in Equation (8.138) under the condition τ = 0: A( ν) =

1 g˜ 2 (− ϕ0ctgν γ 0 ) ⋅ J n2 ϕ 0 ctg γ 0 v

{

2 ∆ω M ωM

[ cT

sin

M

2 πh sin( γ 0 −

ν ) ϕ 0 ctg γ 0

}.

(8.168)

]

The value of ∆ϕ ϕ given by Equation (8.167) deﬁnes a shift in the average value of the transformed target return signal phase with respect to the directional diagram axis for transmission and reception caused by various factors: the asymmetrical directional diagram; effect of the speciﬁc effective scattering area of the underlying surface of a two-dimensional target; inﬂuence of the function J n2 ( x), and changes in the radar range within the directional diagram. If we assume that the function g˜ v2 ( ψ ) can be approximated by the Gaussian law with an effective width ∆v and average angle γ0,25,26 and that the speciﬁc effective scattering area S°(ψ) can be approximated by the exponent,27 and if we take into consideration the condition Tr << TM , as in the opposite case it is possible to measure the phase ϕ, based on Equation (8.167) and in terms of Equation (8.168), we can write ∆ϕ sh = ∆ϕ = ϕ av − ϕ 0 = −ϕ 0 ⋅

[ J n2 ( Mρ )] (1) ∆(v2 ) k1 + ctg γ 0 + ⋅ ctg γ 0 . dn ( ThM ) 4π

{

}

(8.169) ϕsh by ∆ρsh in Equation (8.169). Because ϕ ≅ ρ, we can replace ∆ϕ Reference to Equation (8.169) shows that we can deﬁne the relative error ∆ρsh ρ0

in measuring the distance between the radar and the scanned twodimensional (surface) target. For instance, at ∆v = 5°, γ0 = 65°, n = 1, k1 = 12.9 (the weak rough sea with the wind velocity within the limits of the ∆ω

∆ρ

interval 0, … ,11 km/h), and ω MM = 1.2 , the component of the error ρ0sh due to the speciﬁc effective scattering area S°(ψ) is equal to 0.47%; the component of the error

∆ρsh ρ0

due to changes in the radar range ρ within the directional

diagram is equal to 0.02%; and the component of the error 2 n

∆ρsh ρ0

function J ( Mρ ) is equal to 0.22%. The component of the error

due to the ∆ρsh ρ0

due to

the function J n2 ( Mρ ) is decreased, approximately by the same law as is the ∆ω

modulation index ω MM . Because all cofactors of the function A(ν) can be considered in Equation (8.168) to be linear functions at low values of ν, except for the term g v2 (− ϕ0ctgν γ 0 ), the limits of variations of the phase ϕi are deﬁned

Copyright 2005 by CRC Press

1598_book.fm Page 370 Monday, October 11, 2004 1:57 PM

370

Signal and Image Processing in Navigational Systems

only by the function gv2 (ψ ). If the function gv2 (ψ ) takes the shape of the Gaussian curve, we can write

g (− 2 v

ν ϕ 0ctg γ 0

)≈e

−π

( ϕ − ϕ 0 )2 ∆(ϕ2 )

,

(8.170)

where ∆ ϕ = nω M Tr = 2 2 πn ⋅

h∆ v cos γ 0 cTM sin 2 γ 0

(8.171)

is the effective bandwidth of the phase characteristic of radar at the frequency 2nωM. The value of ∆ϕ can be used to estimate the variance of the ﬂuctuations of the phase ϕ (or radar range ρ) caused by the limited width of the directional diagram σ ϕ2 =

∆(ϕ2 ) 2π

or

For instance, at ∆ν = 1° and γ0 = 65°, we get

8.7

σ ρ2 = σρ ρ0

c 2Tr2 . 8π

(8.172)

= 1.27%.

Conclusions

The fact that the instantaneous power spectral density of the target return signal ﬂuctuations is a function of time allows us to isolate information regarding pulsed target return signal delay, which deﬁnes the distance between the radar and the scanned two-dimensional (surface) target. As is well known, in the use of radar with frequency-modulated searching signals, information about the distance between the radar and the two-dimensional (surface) target is extracted from the frequency of the transformed target return signal. To obtain the frequency of the transformed target return signal from different targets, we use a set of ﬁlters. The bandwidth of each ﬁlter is approximately equal to the modulation frequency. The comb characteristics of these ﬁlters are analogous to pulses gated in time under the use of radar with pulsed searching signals. Because of this, the correlation function of the ﬂuctuations is of prime interest to us and there is no need to consider the instantaneous correlation function. Thus, the correlation function of the transformed ﬂuctuations must be averaged over the period of frequency

Copyright 2005 by CRC Press

1598_book.fm Page 371 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

371

modulation. The general formula for deﬁnition of the correlation function of the ﬂuctuations is given by Equation (8.14). The correlation function averaged over time is the sum of the correlation functions given by Equation (8.14) with speciﬁc coefﬁcients for each interval. With linear variation of the searching signal frequency, the correlation function of the ﬂuctuations is a nonstationary frequency-modulated process. Thus, the Doppler frequency is a function of time. In addition to the Doppler shift in frequency, the target return signal is shifted in frequency by the rangeﬁnder frequency. The range-ﬁnder frequency is shifted by its own Doppler frequency, too. Relative shifts in the range-ﬁnder- and Doppler frequencies with respect to their differences within the directional diagram compensate each other under a ﬁxed aspect angle. If the Doppler frequency is negative, the rangeﬁnder frequency must be positive to ensure power spectral density compression of the ﬂuctuations under sloping scanning of the underlying surface of a two-dimensional target. It should be noted that complete compression of the power spectral density can take place not only with horizontally moving radar, but in all cases when the velocity vector of moving radar and the directional diagram axis are in the same vertical plane. The Doppler frequency as a function of time is present in the transformed target return signal and can hide the effect of power spectral density compression. With the asymmetric saw-tooth frequency-modulated searching signal, the coefﬁcients of the Fourier-series expansion of the correlation function of the ﬂuctuations take the form of the function sinc2 x and the average frequency ωav = ω0 + 0.5∆ωM. The power spectral density consists of a set of partial power spectral densities deﬁned by harmonics of the modulation frequency ωM. Unlike the case of the nonperiodic searching signal, the correlation function of the ﬂuctuations possesses two range-ﬁnder frequencies given by Equation (8.43), which are deﬁned by different intervals of the modulation period. Harmonics of the modulation frequency ωM are shifted by the onehalf Doppler frequency. Because all components of the target return signal must be shifted by their Doppler frequency, not the one-half Doppler frequency, we can reason that another one-half Doppler shift in frequency is given by Equation (8.39). As the condition nωM << ω0 is satisﬁed, as a rule, we can neglect the Doppler shift in modulation frequency ωM. The same conclusion can be reached in the case of the symmetric saw-tooth frequencymodulated searching signal. The partial power spectral densities of the ﬂuctuations with the harmonic frequency-modulated searching signal, as well as with the symmetric sawtooth frequency-modulated searching signal, are at the total and difference frequencies Ωn. There is a narrowing of the power spectral density at the difference frequency and an expansion at the total frequency.

Copyright 2005 by CRC Press

1598_book.fm Page 372 Monday, October 11, 2004 1:57 PM

372

Signal and Image Processing in Navigational Systems

References 1. Scharf, L., Statistical Signal Processing, Detection, Estimation, and Time Series Analysis, Addison-Wesley, Reading, MA, 1991. 2. Li, J., Liu, G., Jiang, N., and Stoica, P., Moving target feature extraction for airborne high-range resolution phased-array radar, IEEE Trans., Vol. SP-49, No. 2, 2001, pp. 277–289. 3. Schleher, D., MTI and Pulsed Doppler Radar, Artech House, Norwood, MA, 1991. 4. Ward, K., Baker, C., and Watts, S., Maritime surveillance radar. Part 1: Radar scattering from the ocean surface, Proceedings of the IEE F, Vol. 137, No. 2, 1990, pp. 51–62. 5. Dong, X., Beaulieu, N., and Wittke, P., Signaling constellations for fading channels, IEEE Trans., Vol. COM-47, No. 5, 1999, pp. 703–714. 6. Rappaport, T., Wireless Communications: Principles and Practice, Prentice Hall, Englewood Cliffs, NJ, 1996. 7. Stuber, G., Principles of Mobile Communications, Kluwer, Boston, MA, 1996. 8. Hudson, J., Adaptive Array Principles, Peter Peregrinus, London, 1981. 9. Muqnet, B., Courville, M., Duhamel, P., and Buzenac, V., A subspace based blind and semi-blind channel identiﬁcation method for OFDM systems, in Proceedings of the IEEE 2nd Workshop on Signal Processing Advances in Wireless Communications, Annapolis, MD, May 9–12, 1999, pp. 170–173. 10. Graziano, V., Propagation correlation at 900 MHz, IEEE Trans., Vol. VT-27, No. 11, 1978, pp. 1182–1189. 11. Jukovsky, A., Onoprienko, E., and Chijov, V., Theoretical Foundations of Radar Altimetry, Soviet Radio, Moscow, 1979. (In Russian.) 12. Jacobs, S. and O’Sullivan, J., Automatic target recognition using sequences of high resolution radar range proﬁles, IEEE Trans., Vol. AES-36, No. 4, 2000, pp. 364–382. 13. Gerlach, K. and Steiner, M., Fast converging adaptive detection of Dopplershifted range-distributed targets, IEEE Trans., Vol. SP-48, No. 9, 2000, pp. 2538–2541. 14. Melvin, W., Wicks, M., Autonik, P., Salama, X., Li, P., and Schuman, H., Knowledge-based space–time adaptive processing for airborne early warning radar, IEEE Aerosp. Electron. Syst. Mag., April 1998, pp. 37–42. 15. Barndoff–Nielsen, O. and Cox, D., Asymptotic Techniques for Use in Statistics, Chapman & Hall, London, 1989. 16. Arnold, H., Cox, D., and Murray, R., Macroscopic diversity performance measured in the 800-MHz portable radio communications environment, IEEE Trans., Vol. AP-36, No. 2, 1988, pp. 277–280. 17. Baker, C., The Numerical Treatment of Integral Equations, Oxford University Press, Oxford, U.K., 1977. 18. Besson, O., Stoica, P., and Gershman, A., Simple and accurate direction of arrival estimator in the case of imperfect spatial coherence, IEEE Trans., Vol. SP-49, No. 4, 2001, pp. 730–737. 19. Amin, M., Time-frequency spectrum analysis and estimation for nonstationary random processes, in Time-Frequency Signal Analysis: Methods and Applications, Longman-Cheshire, London, 1992.

Copyright 2005 by CRC Press

1598_book.fm Page 373 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Two-Dimensional (Surface) Target

373

20. Tyapkin, L., Spectral characteristics of frequency-modulated target return signals, Problems in Radio Electronics, Vol. OT, No. 4, 1976, pp. 3–23. (In Russian.) 21. Ho, P. and Lane, P., Spectrum, distance, and receiver complexity of encoded continuous phase modulation, IEEE Trans., Vol. IT-34, No. 9, 1988, pp. 1021–1032. 22. Sari, H., Orthogonal frequency-division multiple access with frequency-hopping and diversity, in Multi-Carrier Spread Spectrum, K. Fazel and G. Fettweis, Eds., Kluwer, Norwell, MA, 1997, pp. 57–68. 23. Tyapkin, L. and Mandurovsky, I., Spectral characteristics of frequency-modulated target return signals, Problems in Radio Electronics, Vol. OT, No. 2, 1970, pp. 35–58. (In Russian.) 24. Kolchinsky, V., Mandurovsky, L., and Konstantinovsky, M., Doppler Devices and Navigational Systems, Soviet Radio, Moscow, 1979. (In Russian.) 25. Narayanan, K. and Stuber, G., Performance of trellis-coded CPM with iterative demodulation and decoding, IEEE Trans., Vol. COM-49, No. 4, 2001, pp. 676–687. 26. Buzzi, S., Lops, M., and Tulino, A., Time-varying narrowband interference rejection in asynchronous multiuser DS/CDMA systems over frequencyselective fading channels, IEEE Trans., Vol. COM-47, No. 10, 1999, pp. 1523–1536. 27. Buzzi, S., Lops, M., and Tulino, A., MMSE RAKE reception for asynchronous DS/CDMA overlay systems and frequency-selective fading channels, in Wireless Personal Communications, Kluwer, Norwell, MA, Vol. 13, June 2000, pp. 295–318.

Copyright 2005 by CRC Press

1598_book.fm Page 375 Monday, October 11, 2004 1:57 PM

9 Fluctuations under Scanning of the ThreeDimensional (Space) Target by the Continuous Signal with a Frequency that Varies with Time

9.1

General Statements

Let us assume that the radar illuminates the three-dimensional (space) target with the continuous searching signal whose amplitude and frequency vary with time [see Equation (2.54)]. At ﬁrst, let us suppose that the three-dimensional (space) target has a limited length in the radar range interval [ρ1, ρ2] and that the boundary values ρ1 and ρ2 are constants. In other words, the boundary values ρ1 and ρ2 are independent of the variables ϕ and ψ within the radar antenna directional diagram. This assumption does not change the physical essence of the considered problem but makes it simple to solve.1–3 If the frequency of the searching signal is a periodic function of time with period TM , then a product of the shifted signals, W(t – 0.5τ) · W*(t + 0.5τ), similar to Equation (2.62), is a periodic function both of the time t and of the shift in time τ. Thus, the correlation function of the ﬂuctuations in the target return signal frequency given by Equation (2.55) can be written in the following form ∞

R∆ρ, ∆ω (t , τ) = p

π

0.5 π

ρ2

∑ ∫ ∫ ∫ g (ϕ, ψ)ρ 2

−2

⋅ e − j∆Ψ n dρ dϕ dψ

(9.1)

n = − ∞ − π − 0.5 π ρ1

if the amplitude of the target return signal is given by Equation (2.72) and the variables ρ, ϕ, and ψ are independent, where = ω (τ − nT ) − 2ω ∆ρc −1 + ∆Ψ ; ∆Ψ n 0 M 0 n ∆Ψ n = Ψ[ t −

2ρ c

+ 0.5(τ − nTM ) −

∆ρ c

] − Ψ[ t − 2cρ − 0.5(τ − nTM ) + ∆ρc ] ;

(9.2) (9.3)

375 Copyright 2005 by CRC Press

1598_book.fm Page 376 Monday, October 11, 2004 1:57 PM

376

Signal and Image Processing in Navigational Systems

p=

PG02 λ 2 S° . 64π 3

(9.4)

Here we retain the cofactor ejω0nTM for the sake of convenience. In the case of the coherent searching signal, this cofactor is equal to one. We assume that the velocity of variations in frequency kω(t) is a very slow function of the time t so that frequency changes can be considered as a linear function within time intervals that can be compared in value with the correlation length τc . Therefore, we can write Ψ ′(t) = Ω(t) = Ω(t0 + t − t0 ) ≅ Ω(t0 ) + kω (t − t0 )

(9.5)

ω(t) = ω 0 + Ω(t) = ω 0 + Ω(t0 ) + kω (t − t0 ) ,

(9.6)

and

where kω(t) can be a slow function of time, in particular, a constant (but not necessarily) low in value function [see Equation (2.54)]. Without any loss of generality, we can assume that t0 = 0 and Ω(t0) = 0. Then we can write Ψ ′(t) = Ω(t) = kω t

and

ω(t) = ω 0 + kω t .

(9.7)

In the case of the moving radar with constant velocity, i.e., ∆ρ = –Vr · τ, reference to Equation (9.2) and Equation (9.3) shows that ∆Ψ n = (µτ − nTM ) ⋅ Ψ ′(t − 2ρc −1 ) = (µτ − nTM ) ⋅ kω (t − 2ρc −1 ) ;

(9.8)

= ω(t) ⋅ (µτ − nT ) + Ω τ − Ω (µτ − nT ) , ∆Ψ n M D ρ M

(9.9)

µ = 1 + 2 Vr (ϕ , ψ )c = µ(ϕ , ψ )

(9.10)

where

is a coefﬁcient that takes into consideration the Doppler distortion in the target return signal frequency — here, µ(ϕ, ψ) is a function of the coordinates ϕ and ψ, so far we have been using only the average value µ = 1 + ΩD = 2ω(t)Vr (ϕ , ψ )c −1 = 4πVr (ϕ , ψ )λ −1 (t) = ΩD (t , ϕ , ψ )

Copyright 2005 by CRC Press

2 Vr 0 c

;

(9.11)

1598_book.fm Page 377 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Three-Dimensional (Space) Target

377

is the Doppler shift in the target return signal frequency that is proportional to the frequency ω(t) and depends on time; Ωρ = 2 kω c −1ρ = kω Td = Ωρ (t , ρ)

(9.12)

is the range-ﬁnder frequency that is equal to the searching signal frequency differential within the limits of the period Td = 2ρc . The frequencies ω(t) and ΩD(t) are periodic functions of time with the period TM. Neglecting the Doppler shift in the range-ﬁnder frequency

2ρ Ωρ c

in Equation (9.9), which is very low in value, we can write = [ω(t) − Ω ] ⋅ (τ − nT ) + Ω τ . ∆Ψ n ρ M D

(9.13)

The formula in Equation (9.13) shows that there are two components. The ﬁrst component depends on the periodic difference ω(t) – Ωρ; the second component, caused by the Doppler shift, monotonically increases from period to period together with the shift τ = nTM. In the case of the periodic frequency-modulated searching signal, the condition of slow variation of the velocity of changes in frequency kω(t) is not always true. For instance, in the case of the linear frequency-modulated searching signal at deﬁnite instants of time, the velocity of variations in frequency kω(t) is a jump-like function. As a rule, these instants of time can be excluded from consideration because kω(t) = const within the main part of the period.4 In the case of the harmonic frequency-modulated searching signal, we can write Ψ(t) = M ⋅ sin ω M t and Ω(t) = Ψ ′(t) = ∆ω M ⋅ cos ω M t ;

(9.14)

ω(t) = ω 0 + ∆ω M ⋅ cos ω M t ;

(9.15)

M=

∆ω M ; ωM

= µω τ + 2 M ⋅ sin 0.5µω τ ⋅ cos[ω (t − 2ρc −1 )]. ∆Ψ n 0 M M

(9.16)

(9.17)

We can easily show that if the conditions Td = 2ρc << TM and τc << TM are satisﬁed — the last condition is true at high deviation — Equation (9.17) can be rewritten in the form of Equation (9.13), where ω(t) is determined by Equation (9.15) and Ωρ is given by Equation (9.12), and ΩD = 2 Vr (ϕ , ψ )c −1 (ω 0 + ∆ω M cos ω M t);

Copyright 2005 by CRC Press

(9.18)

1598_book.fm Page 378 Monday, October 11, 2004 1:57 PM

378

Signal and Image Processing in Navigational Systems kω (t) = − ∆ω M ⋅ ω M sin ω M t.

(9.19)

If the target return signal was transformed by the product of the heterodyne signal with varying frequency, we must replace the parameter ∆Ψn with ∆Ψn 0 in Equation (9.1)–Equation (9.3), where ∆Ψ n0 = Ψ[t + 0.5(τ − nTM )] − Ψ[t − 0.5(τ − nTM )].

(9.20)

Then, with the previously mentioned assumptions and based on the results discussed in Section 8.3, we obtain = (ω + Ω )τ − Ω (τ − nT ) ∆Ψ n im D ρ M

(9.21)

instead of Equation (9.13). In the case of the harmonic frequency-modulated , method is searching signal, we can ﬁnd an exact formula to deﬁne Α Ψ n cumbersome. Therefore, we use the approximation = [ω + Ω (t)]τ + 4 M ⋅ sin 0.5ω τ ⋅ sin ω M ρ ⋅ sin[ω (t − ∆Ψ M n im D M c

ρ c

)]

(9.22)

when the conditions Td << TM, τc << TM, and 2 cVr << 1 are true. We can also use Equation (9.21) when ΩD is determined by Equation (9.18) and Equation (9.19), Ωρ by Equation (9.12), and kω(t) by Equation (9.18) and Equation (9.19).

9.2

The Nontransformed Target Return Signal

9.2.1

The Searching Signal with Varying Nonperiodic Frequency

In the case of the nonperiodic searching signal, we can use the formulae in Section 9.1 under the condition n = 0. From Equation (9.1) and Equation (9.13) it follows that the normalized correlation function of the target return signal ﬂuctuations has the following form: R∆ρ, ∆ω (t , τ) = Rg (t , τ) ⋅ Rω (t , τ),

(9.23)

where Rg (t , τ) = N

Copyright 2005 by CRC Press

∫∫ g (ϕ, ψ) ⋅ e 2

jΩD ( t ,ϕ , ψ ) τ

dϕ dψ

(9.24)

1598_book.fm Page 379 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Three-Dimensional (Space) Target

379

is the normalized correlation function of the Doppler ﬂuctuations in the target return signal frequency; Rω (t , τ) = Rω° (t , τ) ⋅ e jω (t )τ ,

(9.25)

where ρ2

∫

− jΩ ( t ,ρ) τ Rω (t , τ) = N ρ−2 ⋅ e ρ dρ

(9.26)

ρ1

is the normalized correlation function of the ﬂuctuations in the target return signal frequency. The target return signal given by the normalized correlation function R∆ρ,∆ω(t, τ) is a nonstationary and nonseparable stochastic process because the variables t and τ are not separable. Consider the Doppler ﬂuctuations in the target return signal frequency. Comparing Equation (9.24) with Equation (3.8) and taking into consideration Equation (9.11), we can see the distinction is that the Doppler frequency ΩD in Equation (9.24) uses the variable frequency ω(t) instead of the constant frequency ω0. Because of this, we can use all the formulae of Chapter 3 in the deﬁnition of the normalized correlation function Rg(t, τ) if the frequency 2ω V Ωmax = c0 = 4 πλV is replaced with Ωmax (t) = 2ω(t)Vc −1 = 4πV λ −1 (t).

(9.27)

Thus, the average Doppler frequency of the power spectral density of the ﬂuctuations in the target return signal frequency depends on the time t. If the velocity vector of moving radar is outside the directional diagram, the bandwidth of the power spectral density of the Doppler ﬂuctuations in the target return signal frequency is independent of the wavelength or frequency [for example, see Equation (3.76)] and so does not vary with time. If the velocity vector of moving radar is within the directional diagram, the bandwidth is inversely proportional to frequency [for example, see Equation (3.123)]. When relative variations in the target return signal frequency are low in value, i.e., if the condition ∆ωM << ω0 is satisﬁed, we can neglect these variations.5 When the radar is stationary, the Doppler ﬂuctuations in the target return signal frequency are absent, i.e., Rg(t, τ) 1, and there are only the ﬂuctuations in frequency that are characteristic of the considered case. The cofactor ejω(t)τ in the normalized correlation function Rω(t, τ) deﬁnes the average frequency of the power spectral density of the ﬂuctuations in the target return signal frequency as a function of the time t and leads us to the nonstationary state of the target return signal — a regular dependence on the time t. Another source of the nonstationary state of the target return signal is the dependence of the velocity of variations in frequency kω(t) on the time t.

Copyright 2005 by CRC Press

1598_book.fm Page 380 Monday, October 11, 2004 1:57 PM

380

Signal and Image Processing in Navigational Systems

Under the condition kω(t) = const, the normalized correlation function R°ω(t, τ) of the ﬂuctuations in the target return signal frequency is independent of the time t. This means that the shape and bandwidth of the power spectral density of the ﬂuctuations in frequency do not vary with time in spite of the fact that the frequency of the power spectral density is variable.6 Reference to Equation (9.26) shows that the instantaneous power spectral density of the ﬂuctuations in frequency takes the following form: ρ2

Sω (ω , t) ≅

∫ρ

−2

⋅ δ[ω(t) − kω ⋅ 2cρ − ω] dρ ≅

ρ1

kω , [ω − ω(t)]2

(9.28)

where ω 1 (t) ≤ ω ≤ ω 2 (t)

and

ω 1,2 (t) = ω 0 + kω (t −

2 ρ1 , 2 c

) = ω(t) − kω ⋅ 2ρc

1, 2

.

(9.29) The power spectral density given by Equation (9.29) is shown in Figure 9.1 and Figure 9.2. At the boundary frequencies, the power spectral density Sω(ω1,2) ≅ ρ1−,22 and shape are independent of the time t. The bandwidth of the power spectral density does not vary with time, too, if the condition kω(t) = const is satisﬁed: ∆Ωρ = ω 2 (t) − ω 1 (t) = 2 kω (ρ2 − ρ1 )c −1 = kω ∆Td .

(9.30)

The average frequency of the power spectral density depends on time: ω av = 0.5[ω 1 (t) + ω 2 (t)] = ω(t) − 2 kω ρ0 c −1 = ω(t) − Ωρ0 ,

(9.31)

∆Td = 2(ρ2 − ρ1 )c −1 ;

(9.32)

ρ0 = 0.5(ρ1 + ρ2 );

(9.33)

Ωρ0 = 2 kω ρc −1 .

(9.34)

where

If ρ1 < ρ2 , then S(ω1) > S(ω2). At kω > 0, we have ω1 > ω2 and at kω < 0, we get ω1 < ω2 (see Figure 9.1 and Figure 9.2). The correlation lengths in time τc and in frequency ∆ωc have the following form:

Copyright 2005 by CRC Press

1598_book.fm Page 381 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Three-Dimensional (Space) Target

381

Sω

ω 2(t )

t ω 1(t )

ω (t ) = ω 0 + k ω t

ω ω2

ω1

ω0

FIGURE 9.1 The instantaneous power spectral density of the ﬂuctuations in the target return signal frequency under scanning of the three-dimensional (space) target with the nonperiodic frequencymodulated searching signal: kω > 0.

τc =

2π 2π 1 = = ; ∆Ωρ kω ∆Td k f ∆Td

∆ω c = kω τ c = 2 π( ∆Td )−1

or

∆fc = ( ∆Td )−1 .

(9.35)

(9.36)

Thus, the greater the length ρ1 – ρ2 of the three-dimensional (space) target, the less is the correlation length in time and frequency. Thus, the correlation length in frequency is independent of kω and is deﬁned by the value of ∆Td as a whole. However, it is possible to draw an incorrect conclusion based on Equation (9.35) and Equation (9.36), i.e., as ρ → ∞ the values of τc and ∆ωc tend to approach zero. This can be explained by an insufﬁciently rigorous deﬁnition of the bandwidth of the power spectral density of the ﬂuctuations in frequency, given by Equation (9.30) as a difference of the boundary frequencies. For a rigorous estimation it is necessary to introduce an effective bandwidth of the power spectral density.7 Here, it is appropriate to recall that τc , which is inversely proportional to the bandwidth of the power spectral density, when multiplied by 0.5τ corresponds to the radar range resolution that can be obtained with the searching signal having the same bandwidth. Reference to Equation (9.35) shows that under the condition k f > ∆Td−2 , we get τc < ∆Td , i.e., the radar range resolution is less than the length of the three-dimensional (space) target.

Copyright 2005 by CRC Press

1598_book.fm Page 382 Monday, October 11, 2004 1:57 PM

382

Signal and Image Processing in Navigational Systems Sω ω 1(t ) t ω 2(t )

ω (t ) = ω 0 + k ω t

ω ω0

ω1

ω2

FIGURE 9.2 The instantaneous power spectral density of the ﬂuctuations in the target return signal frequency under scanning of the three-dimensional (space) target with the nonperiodic frequencymodulated searching signal: kω < 0.

In the case of stationary radar, the ﬂuctuations in the target return signal frequency are completely deﬁned by the normalized correlation function given by Equation (9.25) and Equation (9.26) and the power spectral density determined by Equation (9.28). When the radar is moving, it is necessary to deﬁne a convolution between the power spectral density given by Equation (9.28) and the power spectral density of the Doppler ﬂuctuations in the target return signal frequency with the purpose of deﬁning the total power spectral density. This leads to a shift in the average frequency of the instantaneous power spectral density of the ﬂuctuations in frequency by the average Doppler frequency ΩD0 (t) =

2 Vr0 ω ( t ) c

and to an increase in the bandwidth of the

power spectral density of the Doppler ﬂuctuations in frequency. If at some instant of time we have ω(t) = kω ρ0Vr−0 1 ,

(9.37)

the total shift in frequency is determined by ΩD – Ωρ = 0. The total power 0 0 spectral density of the Doppler ﬂuctuations and the ﬂuctuations in the target return signal frequency obtained as a result of convolution of the power

Copyright 2005 by CRC Press

1598_book.fm Page 383 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Three-Dimensional (Space) Target

383

spectral density of the Doppler ﬂuctuations and the power spectral density of the ﬂuctuations in the target return signal frequency is always wider than the widest of that. The total power spectral density compression of the ﬂuctuations in the target return signal frequency is absent, as in the case of scanning the two-dimensional (surface) target. This can be explained by the independence between the variable ρ and the coordinates ϕ and ψ. In the case of the ﬁxed Doppler frequency, i.e., with ﬁxed coordinates ϕ and ψ, there is a scattering over ρ — the power spectral density of the ﬂuctuations in the range-ﬁnder frequency is formed and vice versa, i.e., with ρ ﬁxed, there is scattering at the coordinates ϕ and ψ — and the power spectral density of the Doppler ﬂuctuations in frequency of the target return signal is formed.8 9.2.2

The Periodic Frequency-Modulated Searching Signal

In this case, Equation (9.23) and Equation (9.24) are true, as before, but instead of Equation (9.25) and Equation (9.26) we can write the normalized correlation function of the ﬂuctuations in frequency in the following form: ∞

Rω (t , τ) = N

ρ2

∑ ∫ρ

−2

⋅e

j[ω ( t ) − kω ⋅

2ρ c

]( τ − nTM )

dρ .

(9.38)

n = − ∞ ρ1

The normalized correlation function Rω(t, τ) given by Equation (9.38) is a periodic function of τ as opposed to the normalized correlation function Rωο (t, τ) given by Equation (9.25) and Equation (9.26). The normalized correlation function Rω(t, τ) given by Equation (9.38) is shown in Figure 9.3 with the assumption that changes in the target return signal power are low in value within the limits of the interval [ρ1, ρ2] so that we can replace the value ρ–2 with the value ρ20 . Because of this, the instantaneous power spectral density of the ﬂuctuations in frequency is linear, with the distance between harmonics equal to ωM. As before, the average frequency of the instantaneous power spectral density is shifted by the range-ﬁnder frequency Ωρ0 =

2 ρ0 kω c

with respect to the frequency ω(t),which

is a periodic function of the time t. As a rule, if the conditions 2 ρ0 c

2 ρ0 c

= Td ≤ TM′ , 0

= Td0 ≤ TM′′ , and |kω′ TM′ |=|kω′′TM′′ |= ∆ω M are satisﬁed, then the condition Ωρ

0

<< ∆ωM is true. The power spectral density bandwidth is determined by Equation (9.30), as before. The target return signal is a periodic nonstationary stochastic process. However, periodic frequency modulation is impossible without periodic variations in the velocity of changes in frequency kω. The bandwidth and shape of the power spectral density periodically vary together with variations in the value of kω . The instantaneous power spectral density in the case of the saw-tooth asymmetric linear frequency-modulated searching signal is Copyright 2005 by CRC Press

1598_book.fm Page 384 Monday, October 11, 2004 1:57 PM

384

Signal and Image Processing in Navigational Systems Rω

τ 0

TM

2TM

FIGURE 9.3 The correlation function of the ﬂuctuations in the target return signal frequency under scanning of the three-dimensional (space) target with the periodic frequency-modulated searching signal when the radar is stationary.

shown in Figure 9.5. The bandwidth and shape of the power spectral density are constant within linear intervals. When the sign of the parameter kω is changed, we get a jump-like shift in the power spectral density with respect to the frequency axis.9 Based on Figure 9.5 we can construct the shape of the instantaneous power spectral density both in the cases of the symmetric and one-sided saw-tooth linear frequency-modulated searching signal and also in the case of the harmonic frequency-modulated searching signal (see Figure 9.6), if the frequency ω(t) is determined by Equation (9.16) and the velocity of variations in frequency kω is given by Equation (9.18) and Equation (9.19). The transition processes with duration ∆Td arise in time Td after the jumplike variation of the velocity of changes in frequency kω , which is accompanied by jump-like variations in shape and in the average frequency of the power spectral density. These processes are not investigated here because we assume that the condition ∆Td << TM is satisﬁed. Due to the periodic behavior of the ﬂuctuations in the target return signal frequency, we can deﬁne the intra- and interperiod ﬂuctuations as in the case of the pulsed searching signal (see Chapter 3). Under the condition τ = nTM , the normalized correlation function Rω(t, τ) is equal to one, i.e., the ﬂuctuations caused by the frequency-modulated searching signal are absent, and the interperiod ﬂuctuations in frequency are reduced to the Doppler ﬂuctuations in frequency, as in the case without frequency modulation of the searching signal. The normalized correlation function Rg(t, τ) of the Doppler ﬂuctuations plays the same role of envelope with respect to the normalized correlation function Rω(t, τ) (see Figure 9.4) as in the case of the pulsed Copyright 2005 by CRC Press

1598_book.fm Page 385 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Three-Dimensional (Space) Target

385

R∆ρ, ∆ω

Rg

τ 0

TM

2TM

FIGURE 9.4 The correlation function of the ﬂuctuations in the target return signal frequency under scanning of the three-dimensional (space) target with the periodic frequency-modulated searching signal when the radar is moving.

searching signal. The interperiod Doppler ﬂuctuations can be observed as the ﬂuctuations of the envelope both at the instants of time ﬁxed with respect to the origin of the modulation period (as in the case of the pulsed searching signal) and at the output of the comb ﬁlter, the central frequency of which is within the limits of the interval [ω1(t), ω2(t)]. Under the condition n = 0, the normalized correlation function Rω(t, τ) given by Equation (9.38) deﬁnes the intraperiod ﬂuctuations. Thus, the normalized correlation function Rω(t, τ) coincides with that given by Equation (9.25) and Equation (9.26). The power spectral density of the intraperiod ﬂuctuations is matched with the power spectral density given by Equation (9.28). The intraperiod ﬂuctuations are similar to the previously mentioned ﬂuctuations in the target return signal frequency when the frequency of the searching signal is not periodic. The correlation length of the intraperiod ﬂuctuations given by Equation (9.35) depends on the parameter kf (ρ1 – ρ1). Intraperiod ﬂuctuations are analogous to the rapid ﬂuctuations in the case of the pulsed searching signal, but a distinction is that their power does not vary within the modulation period and the intraperiod ﬂuctuations cannot be caused by propagation of the pulsed searching signal along the threedimensional (space) or two-dimensional (surface) target — the updated set of scatterers in the pulse volume — because at each instant of the time, the target return signal contains elementary signals from all scatterers within the limits of the interval [ρ1, ρ2]. In this case, the intraperiod ﬂuctuations are caused by an updated set of target return signal frequencies, and there is an analogy with the stochastic process observed at the output of the ﬁlter having Copyright 2005 by CRC Press

1598_book.fm Page 386 Monday, October 11, 2004 1:57 PM

386

Signal and Image Processing in Navigational Systems ω (t )

t

Sω

t

TM

ω (t )

ω ω2

ω1

∆ωM ω0 − 2

ω0

FIGURE 9.5 The instantaneous power spectral density of the ﬂuctuations in the target return signal frequency under scanning of the three-dimensional (space) target with the asymmetric linear frequencymodulated searching signal when the radar is stationary.

the frequency response given by Equation (9.28) if the process at the ﬁlter input is white Gaussian noise.10 In the case of moving radar, the instantaneous power spectral density is a convolution of the power spectral densities of the Doppler ﬂuctuations and the ﬂuctuations in the target return signal frequency. As a result, each harmonic of the regular power spectral density Sω(ω, t) has the power spectral density Sg(ω, t) and the total power spectral density shifted by the average Doppler frequency

Copyright 2005 by CRC Press

1598_book.fm Page 387 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Three-Dimensional (Space) Target

387

ω (t )

t

Sω

t TM

ω (t )

ω ω0 ω1 ω2 FIGURE 9.6 The instantaneous power spectral density of the ﬂuctuations in the target return signal frequency under scanning of the three-dimensional (space) target with the harmonic frequency-modulated searching signal when the radar is stationary.

ΩD0 = 2ω(t)Vr0 c −1 .

(9.39)

If the bandwidth of the power spectral density of the Doppler ﬂuctuations is less than the modulation frequency, i.e., if the condition ∆FD < fM is satisﬁed, the instantaneous power spectral density becomes a comb, with clearly detected teeth and with envelope Sω(ω, t) (see Figure 9.7). Under the condition ∆FD > fM , the power spectral density of the ﬂuctuations in the target return signal frequency becomes continuous (see Figure 9.8).

Copyright 2005 by CRC Press

1598_book.fm Page 388 Monday, October 11, 2004 1:57 PM

388

Signal and Image Processing in Navigational Systems

ωM

ω2

ωav

ω1

FIGURE 9.7 The instantaneous power spectral density of the ﬂuctuations in the target return signal frequency under scanning of the three-dimensional (space) target with the periodic linear frequencymodulated searching signal when the radar is moving: ∆FD < fM .

ωM

ω2

ω av

ω1

FIGURE 9.8 The instantaneous power spectral density of the ﬂuctuations in the target return signal frequency under scanning of the three-dimensional (space) target with the periodic linear frequencymodulated searching signal when the radar is moving: ∆FD > fM .

9.2.3

The Average Power Spectral Density with the Periodic FrequencyModulated Searching Signal

We can average the correlation function of the ﬂuctuations in the target return signal frequency using the Fourier-series expansion for the target return signals and multiplying them, and averaging over time without determining the instantaneous correlation function (see Chapter 8). With this approach, we unfortunately lose very important information regarding the instantaneous correlation function and power spectral density. In this section, we deﬁne the average normalized correlation function of the ﬂuctuations in the target return signal frequency by averaging over time the instantaneous correlation function. Let us consider the case of an asymmetric linear frequency-modulated searching signal and assume that kω′ and TM′ are the velocity of frequency Copyright 2005 by CRC Press

1598_book.fm Page 389 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Three-Dimensional (Space) Target

389

variations and duration of the modulation function front, and that kω″ and TM″ are the velocity of frequency variations and duration of the modulation function trailing edge. Integrating the total normalized correlation function R∆ρ,∆ω(t, τ), which is deﬁned by the product of Rg(t, τ) given by Equation (9.24) and Rω(t, τ) given by Equation (9.38), with respect to the variable t within the limits of the period TM = TM′ + TM″ , we obtain the average correlation function of the ﬂuctuations in the target return signal frequency.11 In the case of stationary radar Rg(t, τ) 1, we average only the cofactor ejω(t)τ in Equation (9.38): ∞

Rω (t , τ) = N ⋅ e jω 0τ

∑ [ TT ′ ⋅ R′ (τ − nT M

n= − ∞

ω

M

)+

M

TM′′ ⋅ Rω′′(τ − nTM ) , TM

]

(9.40)

where ρ2

− jkω ′ ⋅ ∆ω M τ Rω′ (τ) = sinc ⋅ ρ−2 ⋅ e 2

∫

2ρ τ c

dρ

ρ1

ρ2

Rω′′(τ) = sinc

∆ω M τ ⋅ ρ−2 ⋅ e 2

∫

an nd ;

2ρ − jkω τ ′′ ⋅ c

(9.41)

dρ

ρ1

∆ω M = |kω′ TM′ | = |kω′′TM′′ | ;

(9.42)

TM = TM′ + TM′′ .

(9.43)

Thus, in the case of stationary radar, Rω (t , τ) is periodic. The average power spectral density is regular. An envelope of the average power spectral density is obtained by convolution of the power spectral density Sω(ω, t) given by Equation (9.28) with the bandwidth given by Equation (9.30) and the rectangle power spectral density with the bandwidth equal to ∆ωM. In the case of the two-sided linear frequency-modulated searching signal, the power spectral density consists of two overlapping regions that are within the limits of the intervals TM′ and TM″ . Their average frequencies are shifted to the right or left with respect to the frequency ω0: Ωρ′ 0 = 2 kω′ ρc −1

(9.44)

Ωρ′′0 = 2 kω′′ρc −1 ,

(9.45)

and

Copyright 2005 by CRC Press

1598_book.fm Page 390 Monday, October 11, 2004 1:57 PM

390

Signal and Image Processing in Navigational Systems

where kω′ and kω″ have different signs. As the condition ∆ωM >> ∆Ωρ is satisﬁed [see Equation (9.30)], an envelope of the power spectral density is very close to the shape of a trapezium with edges having width ∆ωM and length ∆Ωρ. In the case of the one-sided linear frequency-modulated searching signal, there is only a single region corresponding to Rω′ (τ) or Rω″ (τ). If the radar is moving, we can assume that the bandwidth of the power spectral density of the Doppler ﬂuctuations is independent of the time t. Then, the average power spectral density is shifted by the average Doppler frequency and the harmonics spaced at frequencies ω0 + nωM are subjected to convolution with the power spectral density of the Doppler ﬂuctuations, and the regular power spectral density is transformed to the comb power spectral density. Let us consider the case of the harmonic frequency-modulated searching signal. Substituting Equation (9.17) in Equation (9.1) and averaging over time t within the limits of the period, the average normalized correlation function of the Doppler ﬂuctuations and the ﬂuctuations in the target return signal frequency can be written in the following form: R∆ρ, ∆ω (t , τ) = N

∫∫ g (ϕ, ψ) ⋅ J [2 M ⋅ sin(0.5µω 2

0

M

τ)] ⋅ e jµω 0τ dϕ dψ .

(9.46)

Using the well-known expansion ∞

J 0 ( z ⋅ sin α) = J 02 (0.5 z) + 2

∑ J (0.5z) ⋅ cos 2nα 2 n

(9.47)

n=1

and taking into consideration that cos x = 0.5(e jx + e − jx ),

(9.48)

we can write ∞

R∆ρ, ∆ω (t , τ) = N

∑ J ( M) ⋅ R 2 n

gn

(τ),

(9.49)

n= − ∞

where Rgn (τ) = e j(ω 0 + nω M )τ

Copyright 2005 by CRC Press

∫∫

g 2 (ϕ , ψ ) ⋅ e

j ( ω 0 + nω M ) τ

2 Vr ( ϕ , ψ ) c

dϕ dψ

(9.50)

1598_book.fm Page 391 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Three-Dimensional (Space) Target

391

is the normalized correlation function of the Doppler ﬂuctuations for the case of the continuous searching signal [see Equation (3.8)] at the frequency ω = ω0 + nωM. Then, the average power spectral density can be determined as follows: ∞

S(ω , t) =

∑

J n2 ( M) ⋅ Sn (ω − ω 0 − nω M ),

(9.51)

n= ∞ − ∞

where Sn(ω – ω0 – nωM) is the power spectral density in the case of moving radar and the simple harmonic searching signal with the frequency ω0 + nωM, discussed in more detail in Section 3.2 and Section 3.3. In the case of stationary radar, the average power spectral density given by Equation (9.51) is regular, as in the case of a point target, as geometric characteristics of the three-dimensional (space) target are absent in Equation (9.51) (see Figure 9.9): ∞

S(ω , t) =

∑ J (M) ⋅ δ(ω − ω 2 n

0

− nω M ) .

(9.52)

n= − ∞

If the radar is moving, each harmonic of the average power spectral density S(ω , t) given by Equation (9.52) has a corresponding Doppler shift. The bandwidth and shape of the power spectral density of the Doppler ﬂuctuations depend on the shape and position of the directional diagram with respect to the velocity vector of moving radar. M = 0.2 M = 0.6

M = 1.0

M = 2.0 M = 3.0 M = 5.0

−1 0 1 −1 0 1

−1 0 1

−2 −1 0 1 2 −4 −3 −2 −1 0 1 2 3 4 −6 −5 −4 −3 −2 −1 0 1 2 3 4 5 n

FIGURE 9.9 The average power spectral density of the ﬂuctuations in the target return signal frequency under scanning of the three-dimensional (space) target with the harmonic frequency-modulated searching signal with various modulation indexes when the radar is stationary.

Copyright 2005 by CRC Press

1598_book.fm Page 392 Monday, October 11, 2004 1:57 PM

392

9.3 9.3.1

Signal and Image Processing in Navigational Systems

The Transformed Target Return Signal Nonperiodic and Periodic Frequency-Modulated Searching Signals

In the case of the transformed target return signal, Equation (9.23) is true, in which Rg(t, τ) as given by Equation (9.24) is the same, and the normalized correlation function Rω(t, τ) given by Equation (9.25) and Equation (9.26) is modiﬁed in the following manner: If the searching signal is nonperiodic we replace the frequency ω(t) with the parameter ωim , and if the searching signal is periodic we use Equation (9.38) instead of Equation (9.25) and Equation (9.26), into which we replace the frequency ω(t) with the frequency ωim , too. The main distinction between periodic and nonperiodic searching signals is the following: As long as the velocity of variations in frequency kω does not vary with time, i.e., as long as kω = const, the average frequency of the power spectral density does not vary in time, too, if the radar is stationary. When the radar is moving, the average frequency of the power spectral density varies moderately, as does the average Doppler frequency ΩD0. The nonstationary state of the ﬂuctuations in target return signal frequency is caused only by changes in the value and sign of kω within the period. In the case of the one-sided saw-tooth linear frequency-modulated searching signal, the instantaneous power spectral density is independent of time within the main part of the period if the condition kω = const is satisﬁed (see Figure 9.10). In the case of the two-sided saw-tooth linear frequency-modulated searching signal, the value and sign of kω and the range-ﬁnder frequency Ωρ exhibit 0 a jump-like variation within the period. The bandwidth and average frequency ωav = ωim – Ωρ of the power spectral density vary in the same manner 0 (see Figure 9.11). When the radar is stationary, a deﬁnition of the power spectral density is deﬁned by these data. In the case of moving radar, there are the nonstationary Doppler ﬂuctuations in the target return signal frequency. The nonstationary state of the Doppler ﬂuctuations in the target return signal frequency is caused by variations of the carrier frequency ω(τ) and leads to proportional changes in the average Doppler frequency. The power spectral density bandwidth of the Doppler ﬂuctuations remains ﬁxed if the velocity vector of moving radar is outside the directional diagram, or varies in accordance with the frequency ω(t) if the velocity vector is aligned along the direction of the directional diagram axis. The shape of the instantaneous power spectral density of the Doppler ﬂuctuations and the ﬂuctuations in the target return signal frequency is the same as in Figure 9.7 and Figure 9.8 for various relationships between the power spectral density bandwidth of the Doppler ﬂuctuations and the modulation frequency ωM. In the case of the harmonic frequency-modulated signal, the normalized correlation function R∆ρ,∆ω(t, τ) is approximately determined by Equation (9.23)–Equation (9.26) under the conditions Td << TM and τc << TM when we Copyright 2005 by CRC Press

1598_book.fm Page 393 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Three-Dimensional (Space) Target

393

Sω

t

Ωρ

TM • 0

ω ω av

ω im

FIGURE 9.10 The instantaneous power spectral density of the ﬂuctuations in the transformed target return signal frequency with the one-sided linear frequency-modulated searching signal when the radar is stationary. t Sω

TM •

T ′M • Ω′ρ

0

Ω″ρ

0

ω ω′av

ω im

ω″av

FIGURE 9.11 The instantaneous power spectral density of the ﬂuctuations in the transformed target return signal frequency with the two-sided linear frequency-modulated searching signal when the radar is stationary.

Copyright 2005 by CRC Press

1598_book.fm Page 394 Monday, October 11, 2004 1:57 PM

394

Signal and Image Processing in Navigational Systems Sω

t

ω av (t ) Ωρ

0

0

ω ω im FIGURE 9.12 Instantaneous power spectral density of the ﬂuctuations in the transformed target return signal frequency with the harmonic frequency-modulated searching signal when the radar is stationary.

replace the frequency ω(t) with the intermediate frequency ωim in Equation (9.25) and Equation (9.26), and the parameters ΩD and kω are given by Equation (9.18) and Equation (9.19). The corresponding instantaneous power spectral density is regular in the case of stationary radar (see Figure 9.12). This power spectral density is similar to that for the cases of the nontransformed target return signal and the harmonic frequency-modulated searching signal (see Figure 9.6). The power spectral density shown in Figure 9.12 is different from that shown in Figure 9.6; the frequency ω(t) is replaced with the intermediate frequency ωim. In the case of moving radar, it is necessary to deﬁne a convolution of the power spectral densities of the frequency ﬂuctuations and the Doppler ﬂuctuations of the target return signal. The resulting power spectral density becomes either a comb or continuous, in accordance with the relationship between ∆FD and fM (see Figure 9.7 and Figure 9.8).

9.3.2

The Average Power Spectral Density with the Periodic FrequencyModulated Searching Signal

As was shown in Section 9.3.1, when the radar is stationary, i.e., when the condition kω = const is satisﬁed, and with the linear frequency-modulated searching signal, the instantaneous power spectral density of the ﬂuctuations Copyright 2005 by CRC Press

1598_book.fm Page 395 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Three-Dimensional (Space) Target

395

ω ω im

ω av

FIGURE 9.13 The average power spectral density of the ﬂuctuations in the transformed target return signal frequency with the one-sided linear frequency-modulated searching signal when the radar is stationary.

in the target return signal frequency is independent of the time t. For this reason, with the one-sided linear frequency-modulated searching signal, the power spectral density averaged over the modulation period coincides with the instantaneous power spectral density and remains regular (see Figure 9.13). In the case of the two-sided linear frequency-modulated searching signal, the average power spectral density (regular, too) is deﬁned by a weighted sum of the instantaneous power spectral densities Sω′ (ω) and Sω″ (ω) (see Figure 9.14): Sω (ω , t) =

TM′ T ′′ ⋅ Sω′ (ω ) + M ⋅ Sω′′ (ω ) , TM TM

(9.53)

where TM is given by Equation (9.43). If the radar is moving and the regular power spectral density is transformed to a comb, as shown in Figure 9.7 and Figure 9.8, the average frequency of the power spectral density becomes a weak periodic function in time due to the Doppler shift in frequency. An envelope of the power spectral density averaged over the modulation period is additionally fuzziﬁed by the value of

2 Vr0 ∆ω M c

. If

2 Vr0 ∆ω M c

is considerably less than the power spectral

density bandwidth of the Doppler ﬂuctuations in the target return signal frequency, we can neglect it. Deﬁne the average power spectral density in the case of the harmonic frequency-modulated searching signal. For this purpose, we use Equation

ω av '

ω im

" ω av

ω

FIGURE 9.14 The average power spectral density of the ﬂuctuations in the transformed target return signal frequency with the two-sided linear frequency-modulated searching signal when the radar is stationary.

Copyright 2005 by CRC Press

1598_book.fm Page 396 Monday, October 11, 2004 1:57 PM

396

Signal and Image Processing in Navigational Systems

(9.22) and average the normalized correlation function R∆ρ,∆ω(t, τ) over the time t: R∆ρ, ∆ω (t , τ) = Rg (τ) ⋅ Rω (t , τ) ⋅ e jωimτ p,

(9.54)

where Rg ( τ ) = N

∫∫

g 2 (ϕ , ψ ) ⋅ e

V (ϕ ,ψ ) −2 jω 0τ r c

dϕ dψ ;

(9.55)

ρ2

∫

ω ρ Rω (t , τ) = N J 0 4 M ⋅ sin Mc sin(0.5ω M τ) ⋅ ρ−2 dρ .

[

]

(9.56)

ρ1

In this approximation, the Doppler ﬂuctuations and the ﬂuctuations in the target return signal frequency are separable. In the case of stationary radar, the normalized correlation function Rω (t , τ) given by Equation (9.56) deﬁnes exactly and completely the ﬂuctuations in the target return signal frequency. Using Equation (9.47), we can write ∞

∑ N (ρ , ρ ) ⋅ e

Rω (t , τ) =

n

1

2

jnω M τ

,

(9.57)

dρ .

(9.58)

n= − ∞

where ρ2

Nn =

∫ J [2 M ⋅ sin 2 n

ω Mρ c

]⋅ ρ

−2

ρ1

Thus, the average power spectral density is regular (the distance between harmonics being ωM) and has the following form: ∞

Sω (ω , t) ≅

∑N

n

⋅ δ(ω − ω 0 − nω M ) .

(9.59)

n= − ∞

The amplitudes of harmonics depend on the modulation index M and the interval [ρ1, ρ2]. Note that amplitudes of harmonics change with n as the distance between the radar and the three-dimensional (space) target varies.

Copyright 2005 by CRC Press

1598_book.fm Page 397 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Three-Dimensional (Space) Target

397

We can deﬁne this function depending on the determination of the coefﬁcients N. For instance, if the condition

2 ρ0 c

<< TM is true, we can write

ρ2

Nn ≅

∫ J [∆ω 2 n

M

]

⋅ 2cρ ⋅ ρ−2 dρ .

(9.60)

ρ1

In the case of moving radar, the resulting power spectral density is deﬁned by convolution of the regular power spectral density Sω (ω , t) given by Equation (9.60) and the power spectral density of the Doppler ﬂuctuations. Each harmonic is transformed into the power spectral density of the Doppler ﬂuctuations in frequency of the target return signal corresponding to the normalized correlation function Rg(t, τ).12

9.4

Conclusions

The average Doppler frequency of the power spectral density of the ﬂuctuations in the target return signal frequency is a function of time. If the velocity vector of moving radar is outside the directional diagram, the power spectral density bandwidth of the Doppler ﬂuctuations is independent of wavelength or frequency and, for this reason, is not a function of time. If the velocity vector of moving radar is within the directional diagram, the power spectral density bandwidth of the Doppler ﬂuctuations is inversely proportional to frequency. As relative variations in frequency are very low in value, i.e., as the condition ∆ωM << ω0 is satisﬁed, we can neglect frequency variations. In the case of stationary radar, the Doppler ﬂuctuations are absent and there are only the ﬂuctuations in the target return signal frequency that are characteristic of the considered case. The cofactor ejω(t)τ in the normalized correlation function Rω(t, τ) deﬁnes the average frequency of the power spectral density as a function of time and leads to nonstationary target return signals — a regular dependence on time. Another source of the nonstationary state of target return signals is the dependence of the velocity of variations in frequency kω on time. If the parameter kω does not vary with time, the normalized correlation function Rωo (t, τ) is independent of time. This means that the shape and bandwidth of the power spectral density are independent of time in spite of the fact that the frequency of the power spectral density is ﬂoating. The total power spectral density of the Doppler ﬂuctuations and the ﬂuctuations in the target return signal frequency obtained as a result of convolution of the power spectral density of the Doppler ﬂuctuations and the

Copyright 2005 by CRC Press

1598_book.fm Page 398 Monday, October 11, 2004 1:57 PM

398

Signal and Image Processing in Navigational Systems

power spectral density of the ﬂuctuations in the target return signal frequency is always wider than the widest of that. The power spectral density compression of the ﬂuctuations in the target return signal frequency is absent, as in the case of scanning the two-dimensional (surface) target. This can be explained by independence between the variable ρ and the coordinates ϕ and ψ. With a ﬁxed Doppler frequency (the ﬁxed coordinates ϕ and ψ), there is scattering over the radar range ρ, i.e., the power spectral density of the ﬂuctuations in the range-ﬁnder frequency of the target return signal is formed. With a ﬁxed radar range ρ, there is scattering at the coordinates ϕ and ψ, i.e., the power spectral density of the Doppler ﬂuctuations is formed. In the case of periodic behavior of the ﬂuctuations in the target return signal frequency, we can deﬁne the intra- and interperiod ﬂuctuations as in the case of the pulsed searching signal. The interperiod Doppler ﬂuctuations of the target return signal can be observed as the ﬂuctuations of the envelope, both at the instants of time ﬁxed with respect to the origin of the modulation period, as in the case of the pulsed searching signal, and at the output of the comb ﬁlter, the central frequency of which is within the limits of the interval [ω1(t), ω2(t)]. The intraperiod ﬂuctuations in the target return signal frequency are analogous to the rapid ﬂuctuations in frequency with the pulsed searching signal, but a distinction is that their power does not vary within the modulation period, and the intraperiod ﬂuctuations cannot be caused by propagation of the pulsed searching signal along the three-dimensional (space) or two-dimensional (surface) target — the updated set of scatterers in the pulse volume — because at each instant of time, the target return signal contains elementary signals from all scatterers within the limits of the interval [ρ1, ρ2]. In the case of the transformed target return signal, when the radar is moving there are the nonstationary Doppler ﬂuctuations in the target return signal frequency. The nonstationary state of the Doppler ﬂuctuations is caused by variations of the carrier frequency and leads to proportional variations in the average Doppler frequency. The power spectral density bandwidth of the Doppler ﬂuctuations in the target return signal frequency remains ﬁxed if the velocity vector of moving radar is outside the directional diagram, or varies in accordance with the frequency ω(t) if the velocity vector is matched with the directional diagram axis. The shape of the instantaneous power spectral density of the Doppler ﬂuctuations and the ﬂuctuations in the target return signal frequency is the same as with the nontransformed target return signal for various relationships between the power spectral density bandwidth of the Doppler ﬂuctuations in the target return signal frequency and the modulation frequency.

Copyright 2005 by CRC Press

1598_book.fm Page 399 Monday, October 11, 2004 1:57 PM

Fluctuations under Scanning of the Three-Dimensional (Space) Target

399

References 1. Blackman, S. and Popoli, R., Design and Analysis of Modern Tracking Systems, Artech House, Boston, MA, 1999. 2. Farina, A., Antenna-Based Signal Processing Techniques for Radar Systems, Artech House, Norwood, MA, 1992. 3. Hughes, P., A high resolution range radar detection strategy, IEEE Trans., Vol. AES-19, No. 5, 1983, pp. 663–667. 4. Li, J., Liu, G., Jiang, N., and Stoica, P., Moving target feature extraction for airborne high-range resolution phased-array radar, IEEE Trans., Vol. SP-49, No. 2, 2001, pp. 277–289. 5. Schleher, D., MTI and Pulsed Doppler Radar, Artech House, Norwood, MA, 1991. 6. Dickey, F., Labitt, M., and Standaher, F., Development of airborne moving target-radar for long range surveillance, IEEE Trans., Vol. AES-27, No. 6, 1991, pp. 959–971. 7. Muirhead, R., Aspects of Multivariate Statistical Theory, John Wiley & Sons, New York, 1982. 8. Farina, A., Scannabieco, F., and Vinelli, F., Target detection and classiﬁcation with very high range resolution radar, in Proceedings of the International Conference on Radar, Versailles, France, April 1989, pp. 20–25. 9. Winitzky, A., Basis of Radar under Continuous Generation of Radio Waves, Soviet Radio, Moscow, 1961 (in Russian). 10. Wehner, D., High Resolution Radar, Artech House, Norwood, MA, 1987. 11. Jacobs, S. and O’Sullivan, J., High resolution radar models for joint tracking and recognition, in Proceedings of the IEEE National Conference on Radar, Syracuse, NY, May 1997, pp. 99–104. 12. Mensa, D., High Resolution Radar Cross-Section Imaging, Artech House, Norwood, MA, 1991.

Copyright 2005 by CRC Press

1598_book.fm Page 401 Monday, October 11, 2004 1:57 PM

10 Fluctuations Caused by Variations in Frequency from Pulse to Pulse

10.1 Three-Dimensional (Space) Target Scanning 10.1.1 Nonperiodic Variations in the Frequency of the Searching Signal Let us consider the ﬂuctuations of the target return signals from the threedimensional (space) target caused by nonperiodic variations in the frequency of the pulsed searching signal from pulse to pulse. A frequency jump ∆ω is 2ρ∆ω

accompanied by variations in the phase of elementary signals c that are random because the distance ρ is a random variable. Variations in the phase of elementary signals can be reproduced exactly by navigational systems if the searching signal possesses a constant frequency, but scatterers have to be moving according to a deﬁnite law, i.e., radial displacements (velocities) would be proportional to the distance between the radar and the scatterer. Using Equation (2.65) and Equation (2.72), the total instantaneous correlation function of frequency ﬂuctuations of the target return signal is deﬁned by the product of the normalized correlation functions of the Doppler frequency ﬂuctuations and the target return signal frequency ﬂuctuations. R∆ρ, ∆ω (t , τ) = Rg (t , τ) ⋅ Rω (t , τ),

(10.1)

where the normalized correlation function Rg(t, τ) of the Doppler frequency ﬂuctuations coincides with Equation (2.94) or, in the case of absence of scanning, with Equation (3.8). Rω(t, τ) is determined by the following form: Rω (t , τ) = N ⋅ e jω (t )τ ∞

×

∑ ∫ P[z − 0.5(τ − nT

M

)+

∆ρ0 c

] ⋅ P∗[z + 0.5(τ − nTM ) − ∆cρ ] ⋅ e j∆ω z dz . 0

n= − ∞

(10.2) 401 Copyright 2005 by CRC Press

1598_book.fm Page 402 Monday, October 11, 2004 1:57 PM

402

Signal and Image Processing in Navigational Systems

Let us assume that the intrapulse frequency modulation of the searching signal is absent, i.e., the conditions P(t) Π(t) and V = const or ∆ρ0 = –Vr τ 0 are satisﬁed, and variations in frequency are deﬁned by the regular function of time ω ( t ) = ω 0 + Ω( t ),

(10.3)

for example, instability in the frequency of the searching signal caused by supply voltages. Then, Rω(t, τ) can be represented as a function of the time shift τ. After this we can deﬁne the power spectral density of ﬂuctuations in frequency of the target return signal. Hereafter, we assume that the frequency of the searching signal is a function that varies slowly. We can consider this function as a linear function within the limits of time intervals, the duration of which is much more then the period Tp so that ∆ω = kω (t) ⋅ τ or

∆ f = k f (t) ⋅ τ .

(10.4)

Then Rω (t , τ) = N ⋅ e jω (t )τ ∞

×

∑∫

P[ z − 0.5(µτ − nTM )] ⋅ P ∗ [ z + 0.5(µτ − nTM )] ⋅ e jkω (t ) zτ dz .

(10.5)

n= − ∞

The normalized correlation function Rω(t, τ) of ﬂuctuations in the frequency of the target return signal given by Equation (10.5) is similar to that determined by Equation (4.66) in the case of the wide vertical-coverage radar antenna directional diagram, i.e., gv 1. For this reason, we can use here some results discussed in Section 4.4. In particular, in the case of the square waveform pulsed searching signals, the results discussed in Section 4.4.4 are true if we replace the parameter c* Ωγ with the parameter – kω , or the parameter ∆Fτ with the parameter – kfτp. However, we do not study here the normalized correlation function Rω(t, τ) of the target return signal frequency ﬂuctuations given by Equation (10.5) because the most important features — the interperiod frequency ﬂuctuations with frequency variations of the searching signal from period to period — are obtained immediately, and the rapid target return signal frequency ﬂuctuations within the limits of the period when τ < Tp and n = 0 are the same as in the case of the stable searching signal frequency. We carry out a complete analysis of Rω(t, τ) only in the case of the Gaussian searching signal, taking into consideration the intra- and interperiod ﬂuctuations in the target return signal frequency:

Copyright 2005 by CRC Press

1598_book.fm Page 403 Monday, October 11, 2004 1:57 PM

Fluctuations Caused by Variations in Frequency from Pulse to Pulse

Rω (t , τ) = e

jω ( t ) τ − 0.5 πk 2f τ 2pτ

∞

⋅

∑e

−

403

π ( µτ − nTp ) 2 τ 2p

.

(10.6)

n= − ∞

Rω(t, τ) is a comb function with teeth having the period Tp′ =

Tp µ

(see Figure

3.5). The teeth have a Gaussian shape with the effective bandwidth equal to

2τ p , as in the case of the pulsed searching signal with stable frequency.

The envelope of the teeth also possesses the Gaussian shape, with the effective width equal to τc =

2 kf τ p

(10.7)

that deﬁnes the correlation length in time of the interperiod ﬂuctuations of the target return signal. The correlation length in frequency is determined by the following form: ∆f c = k f τ c = 2 τ −p1 .

(10.8)

These values coincide with analogous values in the case of the continuous searching signal if we suppose that τp Tr and have a similar physical meaning. The instantaneous power spectral density of ﬂuctuations in the target return signal frequency is a comb function and similar to that given by Equation (3.150) (see Figure 10.1). The teeth are Gaussian with the effective bandwidth equal to ∆Fω = ( 2 ) −1 k f τ p

(10.9)

and characterize the power spectral density of the target return signal frequency ﬂuctuations. Speaking rigorously, we have to use the absolute value |kω| instead of the value kω because the power spectral density bandwidth is independent of the sign of the parameter kω . The envelope of the teeth is also Gaussian. The bandwidth is deﬁned by the pulse duration and is equal to ∆Fp =

1 2 τp

. The average frequency of the power spectral density ω(t) = ω0

+ Ω0(t) is a function of the time:

Copyright 2005 by CRC Press

1598_book.fm Page 404 Monday, October 11, 2004 1:57 PM

404

Signal and Image Processing in Navigational Systems

S (ω, t )

ω

ω

ω

ω

ω

ω0 + k ω t t ω0

FIGURE 10.1 The instantaneous power spectral density of ﬂuctuations in the target return signal frequency with linear variations in frequency from pulse to pulse.

Sω (ω , t) ≅ e

−

π ω − ω 0 − Ω ( t )  ∆Ω2p

2

∞

⋅

∑e

−

π [ ω − ω 0 − Ω(( t ) − nΩ′p ] 2 ∆Ωω

2

.

(10.10)

n= − ∞

The total normalized correlation function R∆ρ, ∆ω(t, τ) of the target return signal frequency ﬂuctuations given by Equation (10.1) is deﬁned by the product of Rω(t, τ) frequency ﬂuctuations determined by Equation (10.6) and Rg(t, τ) of the Doppler ﬂuctuations, which depends on the shape of the directional diagram and the orientation of its axis relative to the velocity vector of the moving radar. The total power spectral density is deﬁned by the convolution between the power spectral densities frequency ﬂuctuations [see Equation (10.10)] and the Doppler frequency ﬂuctuations. Thus, the total power spectral density is shifted by the average Doppler frequency Ω 0 = 2ω( t ) Vr0 c −1

(10.11)

that is proportional to the carrier frequency ω(t) and, for this reason, depends on the time t by the same law as the carrier frequency. When the directional diagram is high deﬂected and Gaussian, the power spectral density of the target return signal frequency ﬂuctuations is also Gaussian, but the effective bandwidth of the teeth increases and becomes equal to ∆F = ∆Fω2 + ∆FD2 ,

(10.12)

where ∆FD is the effective bandwidth of the power spectral density of Doppler ﬂuctuations in the target return signal frequency [see Equation (3.81)]. Copyright 2005 by CRC Press

1598_book.fm Page 405 Monday, October 11, 2004 1:57 PM

Fluctuations Caused by Variations in Frequency from Pulse to Pulse

405

It should be pointed out that the effective bandwidth of the power spectral density of Doppler ﬂuctuations in the target return signal frequency is independent of the frequency in the case of the high-deﬂected directional diagram (see Section 3.2.1). Because of this, the effective bandwidth is not varied in time, as the dependence on the wavelength λ has disappeared in Equation (3.81): ∆FD ≅ ∆aλ–1 ≅ d a−1 . When the directional diagram is not deﬂected, we get a linear function of the wavelength λ if Equation (3.123) is true for ∆FD and we get a function of the time t.

10.1.2 The Interperiod Fluctuations In the case of the glancing radar range, assuming τ = we can write that

∫

Rω ( t , τ ) = N ⋅ e jω (t )τ Π 2 ( z ) ⋅ e

jkω ( t ) zτ

nTp µ

in Equation (10.5),

dz .

(10.13)

Based on Equation (10.13) after using the Fourier transform, we can write the instantaneous power spectral density of ﬂuctuations in the target return signal frequency in the following form: Sω (ω , t) ≅ Π2

[

ω −ω (t ) kω ( t )

].

(10.14)

Reference to Equation (10.14) shows that the power spectral density Sω(ω, t) coincides in shape with the squared pulsed searching signal. The main parameters of Sω(ω, t) — the average frequency ω(t) and the effective bandwidth ∆F = 0.5π −1kω ( t )τ (p2)

(10.15)

— vary slowly in time. If kω = const, the effective bandwidth of Sω(ω, t) is independent of time. It follows from Equation (10.14) that in the case of the Gaussian pulsed searching signal, the Sω(ω, t) can be written in the following form:

Sω (ω , t) ≅ e

−π

[ ω − ω ( t )]

2

[ 2 π∆F( t )]

2

,

(10.16)

where ∆F(t ) = ( 2 ) −1 k f (t )τp

Copyright 2005 by CRC Press

and k f = 0.5π −1 kω .

(10.17)

1598_book.fm Page 406 Monday, October 11, 2004 1:57 PM

406

Signal and Image Processing in Navigational Systems

S(ω, t )

k ωTp

ω ω (t ) k ωτp FIGURE 10.2 The average power spectral density of interperiod ﬂuctuations in the target return signal frequency in the case of the square waveform pulsed linear frequency-modulated searching signal.

In the case of the square waveform pulsed searching signal, the power spectral density of ﬂuctuations in the target return signal frequency takes the square waveform and has the effective bandwidth equal to (Figure 10.2): ∆F ( t ) = k f ( t )τ p .

(10.18)

The normalized correlation function Rω(t, τ) takes the following form: Rω ( t , τ ) = sinc [0.5kω ( t )τ pτ ] ⋅ e jω (t )τ . The correlation length in time is equal to τ c = frequency is equal to ∆fc = k f τ c =

1 τp

1 kf τ p

(10.19)

. The correlation length in

, i.e., it is 2 times less than in the case

of the Gaussian pulsed searching signal. In the ﬁxed radar range τ = nTp , based on Equation (10.5) we can write

∫

Rω ( t , τ ) = N ⋅ e jω (t )τ Π ( z − 0.5∆ρ 0 ) ⋅ Π ( z + 0.5∆ρ 0 ) ⋅ e

jkω ( t ) zτ

dz . (10.20)

In the case of the Gaussian pulsed searching signal, we can write 2

Rω ( t , τ ) = e − π (∆Fω + ∆F′ Copyright 2005 by CRC Press

2

)τ 2 + jω (t )τ

,

(10.21)

1598_book.fm Page 407 Monday, October 11, 2004 1:57 PM

Fluctuations Caused by Variations in Frequency from Pulse to Pulse

407

where ∆Fω = ( 2 ) −1 k f τ p

(10.22)

is the effective bandwidth of the power spectral density of ﬂuctuations in the target return signal frequency caused by the exchange of scatterers [compare with Equation (3.163) at D = 0]. The duration of the pulsed searching signal is long, the ﬂuctuations in the target return signal frequency caused by the exchange of scatterers are weak, and the impact of variations in frequency is strong. As the duration of the pulsed searching signal is decreased, the stimulus of the unstable state of frequency is decreased, and the stimulus of exchange of scatterers is increased.1 In the case of the square waveform searching signal, we can write

Rω (t , τ) = e

jω ( t ) τ

⋅

[

(

sin 0.5 kω τ p τ 1 − 0.5 kω τ p τ

)]

τ 2 Vr0|| cτ p

.

(10.23)

Comparing Equation (10.23) and Equation (3.31), we can see that Rω(t, τ) given by Equation (10.23) deﬁnes the ﬂuctuations of the stochastic process formed by superposition of the square waveform pulsed searching signals with the duration determined by cτ p 2Vr0

=

cτ p 2 V cosβ 0 cos γ 0

(10.24)

with the linear frequency modulation with deviation kω(t)τp. The normalized correlation function Rω(t, τ) of ﬂuctuations in the target return signal frequency is shown in Figure 3.5 and Figure 3.6. The instantaneous power spectral density of ﬂuctuations in the target return signal frequency is determined by Equation (3.32), in which we have to replace the parameter ∆ωM′ with the parameter kω(t)τp , and the parameter τp′ with the parameter

cτ p 2Vr

.

0

10.1.3 The Average Power Spectral Density Let us consider the power spectral density of ﬂuctuations of the target return signal averaged within the limits of the interval [0, Tp]. If the velocity of variations in the frequency kω(t) is not varied in time, or if we can neglect it within the limits of the interval [0, Tp], then the averaging of the power spectral density of the target return signal ﬂuctuations is not a difﬁcult problem. Reference to Equation (10.13) shows that the only cofactor ejω(t)τ in

Copyright 2005 by CRC Press

1598_book.fm Page 408 Monday, October 11, 2004 1:57 PM

408

Signal and Image Processing in Navigational Systems

Equation (10.13) is averaged within the limits of the interval [t – 0.5Tp , t + 0.5Tp]. Under the linear frequency modulation law for the searching signal, we can write e jω (t )τ = e jω (t )τ ⋅ sinc ( 0.5kωTpτ ) .

(10.25)

The corresponding power spectral density of ﬂuctuations in the frequency of the target return signal has a square waveform with the effective bandwidth equal to ∆F = kfTp . The average normalized correlation function of interperiod ﬂuctuations in the target return signal frequency is deﬁned based on Equation (10.13), Equations (10.19)–(10.21), and Equation (10.23) after replacing the cofactor ejω(t)τ with Equation (10.25). The average power spectral density of interperiod ﬂuctuations in the target return signal frequency is deﬁned by the convolution of the instantaneous power spectral densities with a square waveform and the effective bandwidth equal to ∆F = kfTp . Let us consider some examples. In the case of the Gaussian pulsed searching signal, the average normalized correlation function Rω ( t , τ ) takes the following form: 2 2

Rω (t , τ) = sinc[0.5 kω Tp τ] ⋅ e π ∆Fω τ where ∆Fω =

kf τ p 2

+ jω ( t ) τ

(10.26)

,

. In this case, the power spectral density of ﬂuctuations in

the frequency of the target return signal is shown in Figure 10.3. As before, the bandwidth of the power spectral density is equal to ∆F = kfTp . The difference from the previous case is in the shape of the front and trailing edges of the power spectral density, the bandwidth of which is determined k τ

by ∆F = kfTp . If the frequency is different by the value ± f4 p from the frequency corresponding to the middle or front, the power spectral density value is reduced from 0.8 to 0.2. If the condition Tp >> τp is satisﬁed, as before, the width of the front and trailing edges is very low by value in comparison with the bandwidth of the power spectral density. Thus, under the condition Tp >> τp , the target return signal having ﬂuctuating parameters does not impact the shape of the average power spectral density of ﬂuctuations in the target return signal frequency. This can be explained by the following fact. Regular variations in the frequency of the searching signal lead to high extension of the power spectral density of ﬂuctuations in the target return signal frequency, and therefore, the power spectral density of slow ﬂuctuations in the target return signal amplitude and phase is very narrow and does not contribute to the total shape of the power spectral density. The regular frequency modulation of the searching Copyright 2005 by CRC Press

1598_book.fm Page 409 Monday, October 11, 2004 1:57 PM

Fluctuations Caused by Variations in Frequency from Pulse to Pulse

409

S(ω, t )

k ωTp

ω ω (t ) FIGURE 10.3 The average power spectral density of interperiod ﬂuctuations in the target return signal frequency in the case of the Gaussian pulsed searching signal.

signal leads us to ﬁnd that power spectral density, in which the phases of its components depend on the frequency by a deﬁnite law. The phases of the components of the power spectral density of ﬂuctuations in the target return signal frequency are random. Due to simultaneous regular and random changes in the frequency of the searching signal, which is characteristic of the nonstationary stochastic process, there are correlation relationships between phases of the individual components. Because the instantaneous and average normalized correlation functions and power spectral densities do not contain the information regarding phase relationships of the target return signals, the nonstationary stochastic processes are not characterized completely. The complete information can be found in the two-dimensional power spectral densities of the target return signal ﬂuctuations (see Section 2.2.2). Fluctuating changes in parameters of the target return signal are deﬁned − 0.5π -1k 2 τ 2 τ 2

ω p in the average by the slowly varying cofactors sinc [0.5kωτpτ] and e normalized correlation functions of ﬂuctuations in the target return signal frequency given by Equation (10.26). The rapid varying cofactor sinc [0.5kωτpτ] deﬁnes regular changes in parameters of the target return signal. This cofactor plays the same role as the cofactor e jω0τ, which always exists in the high-frequency correlation function of ﬂuctuations in the target return signal frequency and deﬁnes rapid regular (sinusoidal) changes in parameters of the target return signal. The cofactor e jωτ does not deﬁne the peculiarities of the target return signal as a stochastic process. In the case considered here, the cofactors sinc [0.5kωτpτ] and e jω(t)τ are considered as functions characterizing the regular changes in parameters of the target return signal and do not deﬁne the peculiarities of the target return signal as a stochastic process.

Copyright 2005 by CRC Press

1598_book.fm Page 410 Monday, October 11, 2004 1:57 PM

410

Signal and Image Processing in Navigational Systems

If the radar is moving and the total normalized correlation function of ﬂuctuations in the target return signal frequency is deﬁned by the product of the normalized correlation functions of Doppler ﬂuctuations in frequency and ﬂuctuations in the target return signal frequency, and each normalized correlation function depends on time, it is necessary to average the product with respect to the time t. However, in the majority of the cases,2 we can consider that only the average Doppler frequency Ω0(t) =

· cos β0 cosγ0.

2ω( t )V c

of the power spectral density of Doppler ﬂuctuations in the target return signal frequency is a function of the time t. The bandwidth and shape of the power spectral density are independent of the time t. Under this condition, the total average power spectral density is deﬁned by the convolution between the average power spectral density of ﬂuctuations in the target return signal frequency and the power spectral density of Doppler ﬂuctuations in the target return signal frequency, and by the shift in the average frequency on the average Doppler frequency. In particular, with the Gaussian pulsed searching signal, the average normalized correlation function taking into consideration Doppler ﬂuctuations in the target return signal frequency is determined by the following form: 2

2

Rω (t , τ) = sinc[0.5 kω Tp τ] ⋅ e − π ( ∆Fω + ∆FD )τ

2

+ j [ ω ( t ) + Ω0 ( t )] τ

(10.27)

instead of Equation (10.26), where ∆FD is given by Equation (3.81). 10.1.4 Periodic Frequency Modulation Let us consider the periodic one-sided linear frequency-modulated searching signal with the period TM >> Tp . Instead of Equation (10.5) we can write ∞

Rω (t , τ) = N ⋅ e jω (t )τ

∞

∑ ∑ ∫ Π[z − 0.5(µτ − nT )] ⋅ Π[z + 0.5((µτ − nT )] p

p

n= − ∞ m= − ∞

× e jkω ( τ− mTM ) z dz . (10.28) With the Gaussian pulsed searching signal, we can write the normalized correlation function Rω(t, τ) of ﬂuctuations in the target return signal frequency in the following form: ∞

Rω (t , τ) = e

jω ( t ) τ

⋅

∞

∑ ∑

n= − ∞ m= − ∞

Copyright 2005 by CRC Press

−0.5 τ

e

2

[ (µτ −τnT ) p

2 p

+ k 2f τ 2p ( τ − mTM )2

] .

(10.29)

1598_book.fm Page 411 Monday, October 11, 2004 1:57 PM

Fluctuations Caused by Variations in Frequency from Pulse to Pulse

411

Rω

1

τp 1.41

1.41 kf τp

τ 0 Tp

TM

FIGURE 10.4 The instantaneous normalized correlation function of ﬂuctuations in the target return signal frequency in the case of the one-sided linear frequency-modulated searching signal.

Rω(t, τ) is shown in Figure 10.4 without the cofactor ejω(t)τ. This function has double periodicity with the periods TM and Tp. The width and shape of the teeth of Rω(t, τ) are the same as in the case of the nonperiodic frequencymodulated searching signal. When the radar is moving, Rω(t, τ) given by Equation (10.28) must be multiplied by the normalized correlation function of Doppler ﬂuctuations in the target return signal frequency given by Equation (3.8). The correlation length of Doppler ﬂuctuations in frequency can be both greater and less than the period TM. If the correlation length is less than the period TM (see Figure 10.4, curve 1), the periodic peculiarities of the target return signal caused by the frequency-modulated searching signal disappear and are hidden completely by random Doppler ﬂuctuations in frequency. The normalized correlation function Rω(t, τ) given by Equation (10.29) is similar to the normalized correlation function of the target return signal caused by the pulsed searching signal and radar antenna scanning (see Figure 5.5b) if there is a double periodicity with the periods Tp and Tsc . The power spectral density of ﬂuctuations in the target return signal frequency corresponding to Rω(t, τ) given by Equation (10.29) is regular with discreteness FM (see Figure 10.5a). The width of the teeth is equal to of the envelope is equal to

2 Tp

kf τ p 2

; the width

. If the radar is moving, it is necessary to carry

out a convolution with the power spectral density of Doppler ﬂuctuations in the target return signal frequency, and thus, each harmonic is transformed into the power spectral density of Doppler ﬂuctuations in the target return signal frequency. When the power spectral density is narrow, i.e., the condition ∆FD < FM is true, the total power spectral density takes the shape of a double comb. When it is wide, i.e., the condition ∆FD > FM is satisﬁed, the double comb structure disappears.3

Copyright 2005 by CRC Press

1598_book.fm Page 412 Monday, October 11, 2004 1:57 PM

412

Signal and Image Processing in Navigational Systems ω

ω0

k ″ω

k ′ω

t

TM

S (ω, t ) k ′ω 2 1

ω ω(t ) (a)

ΩM

Ωp

S (ω, t ) ′ k ″ω > k ω 2 1

ω ω(t ) (b) FIGURE 10.5 The instantaneous power spectral density of ﬂuctuations in the target return signal frequency in the case of the two-sided asymmetric linear frequency-modulated searching signal: a) stationary radar, b) moving radar.

The nonstationary state of the target return signal manifests itself as the dependence of the average frequency and the Doppler shift in frequency of the power spectral density of ﬂuctuations in the target return signal frequency on time and as a dependence of the bandwidth of the power spectral density of ﬂuctuations in the target return signal frequency on the parameter kω if the parameter kω is a function of the time. Comparing Figure 10.5a and Figure 10.5b, we can see that the widths of the teeth are different with variations in the absolute value of the parameter kω. With Copyright 2005 by CRC Press

1598_book.fm Page 413 Monday, October 11, 2004 1:57 PM

Fluctuations Caused by Variations in Frequency from Pulse to Pulse

413

the symmetric saw-tooth frequency-modulated searching signal, i.e., when the condition kω′ = kω″ is satisﬁed, the nonstationary state of the target return signal appears as periodic changes in frequency. The shape and bandwidth of the power spectral density are not varied in time within the limits of the modulation period because they are independent of sign of the parameter kω. As kω → ∞, the two-sided saw-tooth linear frequency-modulated searching signal is transformed into the one-sided one and the instantaneous regular power spectral density of ﬂuctuations in the target return signal frequency with unlimited bandwidth (the instantaneous regular white noise) is formed at the instant of the jump in frequency. We can deﬁne the instantaneous power spectral density of ﬂuctuations in the target return signal frequency in the case of the harmonic frequency-modulated searching signal. In this case, we have to use the following formulae: ω(t) = ω 0 + ∆ω M ⋅ sin ω M t

and

kω = ω M ⋅ ∆ω M cos ω M t . (10.30)

The corresponding instantaneous power spectral density is shown in Figure 10.6 at various instants of time. At kω = 0, each tooth of the power spectral density consisting from some δ-functions under the condition kω ≠ 0 is contracted into a single δ-function. In conclusion, we pay attention to the following circumstance. With the pulsed searching signal, the radar heterodyne frequency is not varied, the range ﬁnder frequency is absent, and the average frequency of the power spectral density of ﬂuctuations in the target return signal frequency is matched with the transmitter frequency shifted on the Doppler frequency. S (ω, t )

ω

ω

ω

ω

ω

t ω0

TM ω (t )

FIGURE 10.6 The instantaneous power spectral density of ﬂuctuations in the target return signal frequency in the case of the harmonic frequency-modulated searching signal.

Copyright 2005 by CRC Press

1598_book.fm Page 414 Monday, October 11, 2004 1:57 PM

414

Signal and Image Processing in Navigational Systems

In some cases, the heterodyne frequency can vary in a jump-like manner simultaneously with the variation in frequency of the frequency-modulated searching signal. Then the intermediate frequency of the transformed target return signal is constant for all elementary signals and we have to assume that ω(t) = ωim = const, but the Doppler shift in the frequency is proportional to ω(t), as before. In this case, if the Doppler shift in the frequency is absent and if the condition kω = const is satisﬁed, the instantaneous power spectral density of ﬂuctuations in the target return signal frequency is independent of the time t and the average power spectral density coincides with the instantaneous power spectral density.

10.2 Two-Dimensional (Surface) Target Scanning Let us consider the ﬂuctuations of target return signals from the two-dimensional (surface) target caused by variations in frequency from the pulsed searching signal to the pulsed searching signal and the moving radar. In this case, using Equation (2.65) and Equation (4.6)–Equation (4.10), we can ﬁnd that the total normalized correlation function of ﬂuctuations in the target return signal frequency, as in the case of Equation (4.6), is determined by the product of the normalized correlation functions in the azimuth and aspect-angle planes R∆ρ, ∆ω (t , τ) = Rβ (t , τ) ⋅ Rγ , ω (t , τ) ,

(10.31)

where the normalized correlation function Rβ(t, τ) in the azimuth plane is given by Equation (4.7) and the normalized correlation function Rγ,ω(t, τ) in the aspect-angle plane is similar to that determined by Equation (4.66) in the case where the intrapulse frequency modulation for the searching signal is absent: ∞

Rγ , ω (t , τ) = N ⋅ e jω (t )τ

∑ ∫ Π[z − 0.5(µτ − nT )] ⋅ Π[z + 0.5(µτ − nT )] p

p

n=−∞

× gv2 (ψ ∗ + c∗ z) ⋅ e

j[ kω ( t )− c∗Ω γ ] zτ

dz . (10.32)

The difference is in the exponential factor with variable frequency and the exponent in the integrand: in Equation (10.32) we can see the sum kω(t) –c*Ωγ instead of the parameter –c*Ωγ in Equation (4.66). The normalized correlation function R γ,ω(t, τ) in the aspect-angle plane given by Equation (10.32), as in Equation (4.66), takes into consideration the Copyright 2005 by CRC Press

1598_book.fm Page 415 Monday, October 11, 2004 1:57 PM

Fluctuations Caused by Variations in Frequency from Pulse to Pulse

415

interperiod target return signal ﬂuctuations caused by variations in the searching signal frequency in addition to those caused by the propagation of the searching signal and those caused by the difference in the Doppler frequencies at the aspect-angle plane. Rγ,ω(t, τ) can be investigated in the same way as the normalized correlation function given by Equation (4.66). Here we discuss only some peculiarities. Let us consider only the interperiod ﬂuctuations of the target return signal in the glancing radar range when the condition τ =

nTp µ

is satisﬁed. The intra-

period ﬂuctuations of the target return signal are the same as in the case of the stable frequency of the searching signal. Then, the normalized correlation function Rγ,ω(t, τ) of ﬂuctuations in the target return signal frequency in the aspect-angle plane can be written in the following form: ∞

Rγ , ω (t , τ) = N ⋅ e jω (t )τ

∑ ∫ Π ( z) ⋅ g ( ψ 2

2 v

∗

+ c∗ z) ⋅ e

j[ kω ( t )− c∗Ω γ ] zτ

dz . (10.33)

n= − ∞

Using Equation (10.33), we can deﬁne the power spectral density of the interperiod target return signal ﬂuctuations, which is similar to that given by Equation (4.68). The difference is the following: we use the sum kω(t) –c*Ω γ in Equation (10.33) instead of the parameter –c*Ω γ in Equation (4.68). In the case of the wide vertical-coverage directional diagram, i.e., the condition ∆v >> ∆p is satisﬁed, the shape of the power spectral density of ﬂuctuations in the target return signal frequency, as in Equation (4.69), is deﬁned only by the shape of pulsed searching signals: Sγ , ω (ω , t) ≅ Π2

[

ω − ω (t ) kω ( t ) − c∗Ω γ

].

(10.34)

The average frequency of the power spectral density Sγ,ω(ω, t) given by Equation (10.34) is a function of the time. The effective bandwidth of Sγ,ω(ω, t) is determined by the following form: ∆Ωω ,τ ( t ) = k pτ p [kω ( t ) − c ∗Ω γ ] = ∆Ωω ( t ) − ∆Ω τ

(10.35)

∆Ωω ( t ) = k pτ p kω ( t )

(10.36)

where

and ∆Ωτ is given by Equation (4.70). Thus, linear variations in frequency from the pulsed searching signal to the pulsed searching signal as a function of the sign of the parameter kω(t) can lead us to both expansion and narrowing of the bandwidth ∆Ωτ of the Copyright 2005 by CRC Press

1598_book.fm Page 416 Monday, October 11, 2004 1:57 PM

416

Signal and Image Processing in Navigational Systems

power spectral density of target return signal ﬂuctuations in the aspect-angle plane, as in the case of the continuous linear frequency-modulated searching signal when the power spectral density of the target return signal ﬂuctuations in the range ﬁnder frequency can be added to or subtracted from that of the Doppler ﬂuctuations (see Section 8.2). If the condition kω(t) = const is satisﬁed, the condition ∆Ωω(t) = const is also true. The bandwidth of the power spectral density of the target return signal ﬂuctuations in the aspect-angle plane is a function of the value of the aspect angle γ* or the time-cross-section t. Dependence on the aspect angle γ* is shown in Figure 4.10 by the solid line. Because of this, the bandwidth ∆Ωω,τ given by Equation (10.18) is also a function of the aspect angle γ* or the time cross section t. Addition of the term ∆Ωω to the term ∆Ωτ can lift (kω(t) < 0) or lower (kω(t) > 0) the bandwidth ∆Ωτ(γ*). Under the condition ∆Ωω = ∆Ωτ or kω = c*Ω γ , the bandwidth ∆Ωω,τ = 0, i.e., the Doppler ﬂuctuations of the target return signal in the aspect-angle plane are completely compensated by ﬂuctuations in the target return signal frequency. This phenomenon takes place when kf =

f V ⋅ cosβ 0 sin3 γ ∗ . h ⋅ cos γ ∗

(10.37)

Because the bandwidth of the power spectral density of the target return signal ﬂuctuations is a nonnegative value, then, under the condition γ ∗ = γ ∗ , in which Equation (10.37) is true, the bandwidth ∆Ωω,τ shows a bend (Figure m , h = 10 000 m, f = 10,000 4.10, dotted line). For example, at β0 = 0°, V = 200 sec MHz, the equality ∆Ωω,τ = 0 is true if the conditions γ* = 45° and kf = 100 MHz sec

are satisﬁed. Consider the total power spectral density of the target return signal ﬂuctuations taking into consideration the ﬂuctuations both in the aspect-angle plane and in frequency, and the interperiod target return signal ﬂuctuations in the azimuth plane for the simplest case when the pulsed searching signal and the directional diagram are Gaussian and the directional diagram is wide. Then the power spectral density is also Gaussian, and is determined by Equation (4.140). The bandwidth of the power spectral density of the target return signal ﬂuctuations is determined by ∆Ω = ∆Ω 2h + ( ∆Ω τ − ∆Ωω ) 2

(10.38)

instead of Equation (4.141). When β0 = 0°, at the ﬁrst approximation we can consider that ∆Ωh = 0 [see Equation (4.119)]. Then Equation (10.38) is transformed into Equation (10.35). For this approximation, the interperiod ﬂuctuations of the target return signal are absent in the cross section Copyright 2005 by CRC Press

1598_book.fm Page 417 Monday, October 11, 2004 1:57 PM

Fluctuations Caused by Variations in Frequency from Pulse to Pulse

S (ω, t )

417

nTp

2Tp Tp t

0 -Tp

t

-2Tp

FIGURE 10.7 The Doppler ﬂuctuations and interperiod ﬂuctuations in the target return signal frequency for different time-cross-sections at β0 = 0°.

γ ∗ = γ ∗ ( t = t) (see Figure 10.7). If β0 ≠ 0°, the bandwidth of the power spectral density of the target return signal ﬂuctuations given by Equation (10.38) is not equal to zero, but reaches a maximum in value equal to ∆Ωh.

10.3 Conclusions Under three-dimensional (space) target scanning, in the case of target return signal ﬂuctuations caused by nonperiodic variations in frequency from pulse to pulse, the total normalized correlation function of the target return signal frequency ﬂuctuations is deﬁned by the product of the normalized correlation functions of the target return signal frequency ﬂuctuations and the Doppler ﬂuctuations, which depends on the shape of the directional diagram and the orientation of its axis relative to the vector of velocity of the moving radar. The total power spectral density of the target return signal frequency ﬂuctuations is deﬁned by the convolution between the power spectral densities frequency ﬂuctuations and the Doppler frequency ﬂuctuations of the target return signal. Thus, the total power spectral density of the target return signal ﬂuctuations is shifted on the average Doppler frequency that is proportional to the carrier frequency of the target return signal and, for this reason, the total power spectral density depends on time by the same law as the carrier frequency. If the radar is moving and the total normalized correlation function of the target return signal frequency ﬂuctuations is deﬁned by the product of the normalized correlation functions of Doppler frequency ﬂuctuations and the target return signal frequency ﬂuctuations, and each normalized correlation function depends on time, it is necessary to average the product with respect to the time. However, in the majority of cases, we can consider that only the average Doppler frequency is a function of the time. The bandwidth and Copyright 2005 by CRC Press

1598_book.fm Page 418 Monday, October 11, 2004 1:57 PM

418

Signal and Image Processing in Navigational Systems

shape of the power spectral density of the Doppler target return signal frequency ﬂuctuations are independent of the time. In this condition, the total average power spectral density of the target return signal frequency ﬂuctuations is deﬁned by the convolution between the average power spectral density and the power spectral density of Doppler ﬂuctuations and by shift of the average frequency by the average Doppler frequency. In the case of target return signal ﬂuctuations caused by variations in frequency from the pulsed searching signal to the pulsed searching signal and the moving radar, under the two-dimensional (surface) target scanning, the total normalized correlation function of the target return signal frequency ﬂuctuations is deﬁned by the product of the normalized correlation functions in the azimuth and aspect-angle planes. That in the aspect-angle plane takes into consideration the interperiod target return signal frequency ﬂuctuations caused by variations in the frequency of the searching signal in addition to the intraperiod target return signal frequency ﬂuctuations caused by the propagation of the searching signal and by a difference in the Doppler frequencies in the aspect-angle plane.

References 1. Wehner, D., High Resolution Radar, Artech House, Norwood, MA, 1987. 2. Mensa, D., High Resolution Radar Cross-Section Imaging, Artech House, Norwood, MA, 1991. 3. Schleher, D., MIT and Pulsed Doppler Radar, Artech House, Norwood, MA, 1991.

Copyright 2005 by CRC Press

1598_book.fm Page 419 Monday, October 11, 2004 1:57 PM

Part II

Generalized Approach to Space–Time Signal and Image Processing in Navigational Systems

Copyright 2005 by CRC Press

1598_book.fm Page 421 Monday, October 11, 2004 1:57 PM

11 Foundations of the Generalized Approach to Signal Processing in Noise

The generalized approach to signal processing in the presence of noise is based on a seemingly abstract idea: the introduction of an additional noise source, which does not carry any information about the signal, for the purpose of improving the detection performance and noise immunity of complex navigational systems. The proposed generalized approach in the presence of noise allows us to formulate decision-making rules based on the determination of the jointly sufﬁcient statistics of the likelihood function or functional mean and variance. Classical and modern signal processing theories allow us to deﬁne only the sufﬁcient statistic of the likelihood function or functional mean. The presence of additional information about the statistical likelihood function or functional parameters leads to better qualitative characteristics of signal detection in comparison to the optimal signal processing algorithms of classical and modern theories.

11.1 Basic Concepts The simplest signal detection problem is that of binary detection in the presence of additive Gaussian noise with zero mean and the power spectral density 0.5 N0. The optimal detector can be realized as the matched ﬁlter or the correlation receiver. Detection quality depends on the normalized distance between two signal points of the decision-making space. This distance is characterized by signal energies, the coefﬁcient of correlation between the signals, and the spectral power density of the additive Gaussian noise. If the signal energies are equal, then the optimal coefﬁcient of correlation is equal to − 1. Further, the signal waveform is of no consequence. Despite the classical signal processing theory being very orderly and smooth, it cannot provide the most complete answer to the questions posed in the following text. Let us consider brieﬂy the results discussed in References 1 to 49.

Copyright 2005 by CRC Press

1598_book.fm Page 422 Monday, October 11, 2004 1:57 PM

422

Signal and Image Processing in Navigational Systems

The hypothesis H0 must be chosen so that the input stochastic process is normal Gaussian and has a zero mean, vs. the alternative H1, which is also normal Gaussian but has a mean that varies according the known law a(t). In a statistical context, this problem is solved as follows. Let X(t) be the input stochastic process, which is observed within the limits of the time interval [0, T]; a(t), the signal; and ξ(t), the additive Gaussian noise with a zero mean and the known variance σ 2n : X(t ) = a (t ) + ξ(t ) ⇒ H 1 and X(t ) = ξ(t ) ⇒ H 0 .

(11.1)

As elements of the observed input stochastic sample, we take the uncorrelated coordinates T

Xi = λi

∫ X (t)Ξ (t) dt,

(11.2)

i

0

where X(t) is the realization of the input stochastic process within the limits of the time interval [0, T], and λi and Ξi are the eigenvalues and eigenfunctions of the integral equation T

∫

F(t ) = λ R( y − t )Ξ( y) dy , 0 < t < T ,

(11.3)

0

where R(t) is the known correlation function of the additive Gaussian noise. As a rule, we take only the ﬁrst N coordinates. Thus, for the hypothesis H0 the likelihood function of the observed input stochastic sample X1, …, XN has the following form (note: for simplicity, we set the noise variance to be equal to unity): N   1 fX|H0 (X |H 0 ) = ⋅ exp − 0.5 Xi2  . 0.5 N (2 π )   i=1

∑

(11.4)

This notation corresponds to a “no” signal in the observed input stochastic sample X1, …, XN. For the observed input stochastic sample with a nonzero mean a(t), for example, when considering the hypothesis H1, we take for observed coordinates the values T

∫

Xi = ai + ξ i = λ i [ a(t) + ξ(t)] Ξi (t) dt , 0

Copyright 2005 by CRC Press

(11.5)

1598_book.fm Page 423 Monday, October 11, 2004 1:57 PM

Foundations of the Generalized Approach to Signal Processing in Noise

423

where ξ(t) is the additive Gaussian noise with a zero mean and known variance. Then the likelihood function in the presence of a signal in the observed input stochastic sample X1, …, XN has the following form: fX|H1 (X |H1 ) =

N  1 ⋅ exp − 0 . 5 Xi − ai  (2 π)0.5 N  i=1

∑{



}  . 2

(11.6)



This notation corresponds to a “yes” signal in the observed input stochastic sample X1, …, XN. Using Equation (11.4) and Equation (11.6), we can write the likelihood function ratio in the following form:

∑ { ∑

N 2  exp − 0.5 Xi − ai  fX|H1 (X |H1 ) = i 1   = N fX|H0 (X |H 0 )  2 exp − 0.5 Xi  i=1  

}

(11.7)

N  N  ai2  = X1 , … , X N = C , Xi ai − 0.5 = exp   i=1  i = 1

∑

∑

(

)

where C is the constant, which is determined by the performance criterion of the decision-making rule. Taking the logarithm in Equation (11.7), we can write N

∑

N

Xi ai > K op ⇒ H1

or

∑

N

∑a ,

Xi ai ≤ K op ⇒ H 0 , K op = ln C + 0.5

i=1

i=1

2 i

i=1

(11.8) where

∑

N i=1

ai2 is the signal energy and Kop is the threshold. Letting N →

∞ and transitioning to the integral form, and using the Parseval theorem,8 we can maintain generality and write T

∫ X(t)a(t) dt > K

op

⇒ H1

or

0

T

∫ X(t)a(t) dt ≤ K 0

where

∫

T

T

op

(11.9)

∫

⇒ H 0 , K op = ln C + 0.5 a2 (t) dt , 0

a 2 ( t ) dt is the signal energy, and [0, T] is the time interval, within

0

the limits of which the input stochastic process is observed. It is asserted Copyright 2005 by CRC Press

1598_book.fm Page 424 Monday, October 11, 2004 1:57 PM

424

Signal and Image Processing in Navigational Systems

that the signal detection algorithm given by Equation (11.8) and Equation (11.9) reduces to the calculation of the value

∑

N i=1

∫

Xi ai or

T

X ( t ) a ( t ) dt and

0

comparison to the threshold Kop. This signal processing algorithm is optimal for any of the following chosen performance criteria: the Bayesian criterion (including, as particular cases, the a posteriori probability distribution density maximum and maximal likelihood), the Neyman–Pearson criterion, and the mini–max criterion, and is called the correlation signal processing algorithm because the mutual correlation function between the input stochastic process X(t) and signal a(t) is deﬁned. Analysis of the signal processing algorithm given by Equation (11.8) and Equation (11.9) yields a property that, in comparison with other factors, deﬁnes the noise immunity. The essence of the analysis reduces to substituting the actual values Xi = ai + ξi and X(t) = a(t) + ξ(t) (the hypothesis H1) or Xi = ξi and X(t) = ξ(t) (the hypothesis H0) into Equation (11.9):

∑

N

N

N

Xi ai =

i=1

∑ ∑ ai2 +

i=1

N

N

ai ξ i ⇒ H1 ;

i=1

∑

Xi ai =

i=1

∑a ξ ⇒ H i i

0

(11.10)

i=1

or T

T

T

∫ X(t)a(t) dt = ∫ a (t) dt + ∫ a(t)ξ(t) dt ⇒ H 2

0

0

0

T

T

0

0

1

; (11.11)

∫ X(t)a(t) dt = ∫ a(t)ξ(t) dt ⇒ H , 0

∑

where the terms

∑

N i=1

ai ξ i and

∫

T

N i=1

ai2 and

∫

T

a 2 ( t ) dt are the signal energy, and the terms

0

a ( t )ξ ( t ) dt are the noise component with a zero mean and

0

the ﬁnite variance deﬁned N

lim

N→∞

or as T → ∞

Copyright 2005 by CRC Press

{∑ a ξ }

2

i i

i=1

= 0.5Ea N0

as N → ∞

(11.12)

1598_book.fm Page 425 Monday, October 11, 2004 1:57 PM

Foundations of the Generalized Approach to Signal Processing in Noise

∞

∞

∞

[∫ a (t)ξ(t) dt] = ∫ dt∫ a (t)a (s )ξ(t)ξ(s ) ds = 0.5E N . 2

a

0

0

The detection parameter q =

425

0

(11.13)

0

2Ea N0

is taken as a qualitative characteristic of

the signal detection algorithm given by Equation (11.8) and Equation (11.9). This parameter may also be called the voltage signal-to-noise ratio (SNR). This parameter is very important and, together with other factors, deﬁnes the noise immunity of any signal processing system and also, in particular, of navigational systems.

11.2 Criticism Let us consider those factors generating questions in the synthesis of the signal processing algorithm given by Equation (11.8) and Equation (11.9). It is known that

∑

N i=1

∑

N i=1

Xi , which is the sufﬁcient statistic of the mean, and

Xi2 , which is the sufﬁcient statistic of the variance, are the jointly suf-

ﬁcient statistics characterizing the distribution law of the random variable Xi.8,19 The sufﬁcient statistics

∑

N i=1

Xi2 of the likelihood functions fX|H1(X|H1)

and fX|H0(X|H0) are reduced in the synthesis of the signal processing algorithm given by Equation (11.8) and Equation (11.9). This is indeed the case in regard to the form of the expressions and assumptions of the statistical decision theory of decision-making. However, in the physical form, it causes a speciﬁc perplexity. The point is that a “yes” signal — the mean ai of the observed input stochastic sample X1, …, XN is not zero — is indicated in the numerator of Equation (11.7), and a “no” signal is indicated in its denominator with the same coordinates. It would be difﬁcult to imagine another approach to the same input stochastic sample X1, …, XN in both the numerator and denominator of the likelihood function ratio. The ﬁrst question that arises is: Might a signal processing algorithm be constructed without loss of sufﬁcient statistic of variance, which is one of the characteristics of the probability distribution density? Another factor generating questions is that the signal processing is performed against the background of the noise component

Copyright 2005 by CRC Press

1598_book.fm Page 426 Monday, October 11, 2004 1:57 PM

426

∑

Signal and Image Processing in Navigational Systems

N i=1

ai ξ i or

∫

T

a ( t )ξ ( t ) dt caused by the interaction between the signal and

0

noise. The variance of the noise component is proportional to the signal energy, which follows from Equation (11.12) and Equation (11.13). The fact that the voltage signal-to-noise ratio for the signal processing algorithm, which is given by Equation (11.8) and Equation (11.9), deﬁned by Equation (11.12), is not proportional to

2Ea N0

but rather to the square root of the value

2Ea N0

is a

consequence of this. The resulting question would be: Is this good or bad? We would believe it is good if the condition

2Ea N0

< 1 is satisﬁed. However, if

the condition is q < 1 and the probability of false alarm PF is equal to 10−3, for example, the probability of detection PD does not exceed 0.1, which is a practically inoperative region for signal detection. If the conditions and q =

2Ea N0

2Ea N0

>1

are satisﬁed, then the probability of detection PD is smaller in

comparison to the proportional dependence q =

Ea N0

. This conclusion seems

unusual, but it is real and is shown in References 50 and 51. Analyzing Equation (11.8) and Equation (11.9), we may note that this signal processing algorithm is considered to be optimal during the following conditions. The likelihood function or functional ratio is formed using the same input stochastic sample where the numerator assumes a “yes” signal and the denominator assumes a “no” signal in the input stochastic process. In this case, the standard quadratic statistic is reduced and the additional information is lost — the sufﬁcient statistic of the variance of the likelihood function or functional. The expression obtained ensures the calculation of the sufﬁcient statistic of the likelihood function or functional mean only:

∑

N i=1

Xi ai or

∫

T

0

X ( t ) a ( t ) dt , where ai or a(t) is the known signal. Theoreti-

cally speaking, the signal processing algorithm given by Equation (11.8) and Equation (11.9) is not realizable for the following reasons: (1) the mutual correlation function between the input stochastic process Xi or X(t) and the signal ai or a(t) is deﬁned by the left side of the Equation (11.8) and Equation (11.9), respectively; (2) the left side of Equation (11.8) and Equation (11.9) vanishes given a “no” signal in the input stochastic process Xi or X(t):

∑

N i=1

Xi ai or

∫

T

0

X ( t ) a ( t ) dt , where ai = 0 or a(t) = 0, and any physical meaning

is lost. In practice, the signal processing algorithm given by Equation (11.8) and Equation (11.9) is realized if the signal structure ai or a(t) is replaced by its model ai* or a*(t) at the receiver, as ai or a(t) is the completely known signal – ai* = kai or a*(t) = ka(t), where k is the coefﬁcient of proportionality Copyright 2005 by CRC Press

1598_book.fm Page 427 Monday, October 11, 2004 1:57 PM

Foundations of the Generalized Approach to Signal Processing in Noise

∑

N i=1

Xi ai∗ or

∫

T

427

X ( t ) a ∗ ( t ) dt. In this case, the left side of Equation (11.8) and

0

Equation (11.9) has a speciﬁc physical form. If the signal structure ai or a(t) is replaced by its model ai* or a*(t), then the noise component

∫

∑

N i=1

ai∗ξ i or

T

a*(t)ξ(t) dt arises, caused by the interaction between the model signal and

0

noise, and always exists independently of what hypothesis (H0 or H1) is considered. The variance of the noise component noted here is proportional to the energy of the model signal, i.e., 0.5Ea* N0 , where Ea* is the energy of the model signal and 0.5N0 is the power spectral density of the noise. The signal processing algorithm given by Equation (11.8) and Equation (11.9) does not allow us to obtain the ratio between the energy characteristics of the signal and noise in the pure form, for example, in the form

2Ea N0

. It causes the

probability of detection PD to be a function of the square root of the signal and noise energy characteristics ratio, i.e., the voltage signal-to-noise ratio is proportional to

2Ea N0

. The signal processing algorithm given by Equation

(11.8) and Equation (11.9) does not afford detection of the signal and deﬁnition of the signal parameters, whose structure does not correspond to that of the model signal at the receiver. In general, a receiver constructed according to the signal processing algorithm given by Equation (11.8) and Equation (11.9) must be a tracker, not a true detector, because the instant of signal’s appearance on the time axis is unknown relative to the origin. Considering the conditions of optimality of the signal processing algorithm given by Equation (11.8) and Equation (11.9) as brieﬂy outlined above, if the same input stochastic sample is observed in the numerator and denominator of the likelihood function or functional ratio, it is the author’s opinion that it is necessary to undertake a critical review of the initial premises constituting the basis of the classical and modern signal processing theories.

11.3 Initial Premises The signal processing algorithm given by Equation (11.8) and Equation (11.9) is based on the assumption that the frequency–time region Z of the noise exists where a signal may be present; for example, there is an observed stochastic sample from this region, relative to which it is necessary to make the decision a “yes” signal (the hypothesis H1) or a “no” signal (the hypothesis H0 ). We now proceed to modify the initial premises of the classical and modern signal processing theories. Let us suppose there are two independent

Copyright 2005 by CRC Press

1598_book.fm Page 428 Monday, October 11, 2004 1:57 PM

428

Signal and Image Processing in Navigational Systems

frequency–time regions Z and Z* belonging to the space A (see Figure 11.1). Noise from these regions obeys the same probability distribution densities with the same statistical parameters. For the sake of simplicity of this analysis, the same probability distribution density and equality of the statistical parameters have been chosen. However, generally, these parameters may not be equal. A “yes” signal is possible in the noise region Z as before. It is known a priori that a “no” signal is obtained in the noise region Z*. In the following text we will call the noise region Z* the reference region, and consequently, the observed sample from this region is called the reference sample. It is necessary to make the decision a “yes” signal (the hypothesis H1) or a “no” signal (the hypothesis H0) in the observed stochastic sample from the region Z, by comparing the probability distribution density statistical parameters of this sample with those of the sample from the reference region Z*. The problem posed in the preceding sections of this chapter must be solved using the statistical decision-making theory. Thus, it is necessary to accumulate and compare statistical data deﬁning the probability distribution density statistical parameters of the observed input stochastic samples from two independent frequency–time regions Z and Z*. If the probability distribution density statistical parameters for the two samples are equal or agree with each other within the limits of a given accuracy, then the decision of a “no” signal in the observed input stochastic sample from the region Z is made — the hypothesis H0 . If the probability distribution density statistical parameters of the observed input stochastic sample from the region Z differ from those of the reference sample from the region Z* by a value that exceeds the prescribed error limit, then the decision of a “yes” signal in the region Z is made — the hypothesis H1 .

11.4 Likelihood Ratio Now the problem is to obtain jointly sufﬁcient statistics to deﬁne the statistical parameters of the probability distribution densities. For this purpose, let us avail ourselves of one of the well-known results.7,23,28,30,44,52–54 It is known that the sufﬁcient statistic is determined from the condition that the likelihood function has an extremum. In general, the condition of an extremum of the likelihood function, relative to the parameter to be determined with the prescribed accuracy, is determined in the following form: ∂f X ( X 1 ,..., X N |ϑ ) = 0, ∂ϑ

Copyright 2005 by CRC Press

(11.14)

X1, …, XN

In Wx (X1, …, XN ) DSS +

+ η1, …, ηN Z*

− DSS In Wη(η1, …, ηN )

DSS – Definition of Sufficient Statistic FIGURE 11.1 The deﬁnition of jointly sufﬁcient statistics.

In

Wx (X1, …, XN ) Wη(η1, …, ηN )

1598_book.fm Page 429 Monday, October 11, 2004 1:57 PM

Z

Foundations of the Generalized Approach to Signal Processing in Noise

Copyright 2005 by CRC Press

A

429

1598_book.fm Page 430 Monday, October 11, 2004 1:57 PM

430

Signal and Image Processing in Navigational Systems

where N is the sample size determining the prescribed accuracy, and ϑ is the parameter to be determined. However, this equation is not used in practice. A simple mathematical procedure simpliﬁes the representation of this equation. Because the logarithm is a monotonic function, the extrema of the functions fX(X1, …, XN) and lnfX(X1, …, XN) are reached at the same values of the parameter ϑ. Therefore, the likelihood function equation is usually written in the following form: ∂ ln fX (X1 , ... , X N |ϑ) =0. ∂ϑ

(11.15)

As was shown in References 19 and 31,19,31 using Equation (11.4), Equation (11.6), and Equation (11.15), it is not difﬁcult to prove that the values

∑

N i=1

Xi ai and

∑

N i=1

Xi2 are the jointly sufﬁcient statistics of the likelihood

function parameters in Equation (11.4) and Equation (11.6) for the observed input stochastic sample X1, …, XN. The likelihood function for the reference sample η1, …, ηN for unit variance is determined in the following form:

f η ( η1 , ... , ηN ) =

where

∑

N i=1

N

∑ }

1 ⋅ exp − 0.5 ηi2 , (2 π)0.5 N i =1

{

(11.16)

ηi2 is the sufﬁcient statistic of the likelihood function parameters

for the reference sample η1, …, ηN. In the deﬁnition of the sufﬁcient statistics using the input stochastic samples X1, …, XN and η1, …, ηN, the problem of their comparison arises. Usually, for this purpose, a difference is used (see Figure 11.1). The resulting sufﬁcient statistics are observed at the output of the difference device: N

N

N

N

{∑ 2X a − ∑ X + ∑ η − ∑ a } .

ln fX (X1 , ... , X N ) − ln f η ( η1 , ... , ηN ) = 0.5

2 i

i i

i=1

i=1

2 i

i=1

2 i

i=1

(11.17) It is customary to reference the last term on the right side of Equation (11.17) to a threshold independent of the observed input stochastic sample, as in Equation (11.8). Equation (11.17), obtained by the deﬁnition of the resulting sufﬁcient statistics, is the logarithm of the likelihood function. The signal processing algorithm based on the two independent observed input stochastic samples, one of which is the reference sample with a priori information of a “no” signal, follows from Equation (11.17)

Copyright 2005 by CRC Press

1598_book.fm Page 431 Monday, October 11, 2004 1:57 PM

Foundations of the Generalized Approach to Signal Processing in Noise

{ ff ((Xη ,, ......,, ηX ))}

ln fX (X1 , ... , X N ) − ln f η ( η1 , ... , ηN ) = ln N

= 0.5

N

N

X

1

η

1

N

N

N

{∑ 2X a − ∑ X + ∑ η − ∑ a } = ln C 2 i

i i

i=1

2 i

i=1

431

(11.18)

2 i

i=1

i=1

or N

∑

N

2 Xi ai −

i=1

∑

N

Xi2 +

i=1

∑η

2 i

= Kg ,

(11.19)

i=1

where Kg is the threshold. Proceeding from generally accepted concepts, it follows that the hypothesis H1 — a “yes” signal in the observed input stochastic sample X1, …, XN — is assumed if the following inequality is satisﬁed: N

∑

N

2 Xi ai −

i=1

∑

N

Xi2 +

i=1

∑η

2 i

> Kg ,

(11.20)

i=1

and the hypothesis H0 — a “no” signal in the observed input stochastic sample X1, …, XN — is assumed if the opposite inequality is satisﬁed. The ﬁrst term on the left side of Equation (11.20) is the signal processing algorithm given by Equation (11.8) and Equation (11.9) with the factor 2. The more rigorous form of Equation (11.20), based on the analysis performed in Sections 11.1 and 11.2, is the following: N

∑

N

2 Xi ai∗ −

i=1

∑

N

Xi2 +

i=1

∑η

2 i

> Kg ,

(11.21)

i=1

where ai* is the model signal. Letting N → ∞ and transitioning to the integral form, and using the Parseval theorem,8 we maintain generality and can write T

∫

T

∗ i

2 X ( t ) a ( t ) dt − 0

∫X 0

T

2

( t ) dt +

∫ η (t) dt > K 2

g

,

(11.22)

0

where [0, T] is the time interval, within the limits of which the input stochastic process is observed. Analysis of the signal processing algorithm in Equation (11.21) and Equation (11.22), performed by the same procedure as in Sections 11.1 and 11.2, shows that when considering the hypothesis H1: Copyright 2005 by CRC Press

1598_book.fm Page 432 Monday, October 11, 2004 1:57 PM

432

Signal and Image Processing in Navigational Systems

Xi = ai + ξi or X(t) = a(t) + ξ(t) given ai = ai* or a*(t), the left side of Equation (11.21) and Equation (11.22) has the following form: N

2

∑

N

[ ai + ξ i ]ai∗ −

i=1

∑

N

∑

[ ai + ξ i ]2 +

i=1

N

ηi2 =

i=1

N

N

∑ ∑ ai2 +

i=1

ηi2 −

i=1

∑ξ

2 i

(11.23)

i=1

or T

T

∫

T

∫

∫ η (t) dt

2 [ a(t) + ξ(t)]a∗ (t) dt − [ a(t) + ξ(t)]2 dt + 0

0

=

0

T

T

,

(11.24)

T

∫ a (t) dt + ∫ η (t) dt − ∫ ξ (t) dt 2

2

respectively, where the terms

∑

2

0

0

and the terms

2

N i=1

∑

ηi2 −

0

∑

N i=1

N i=1

ai2 and

ξ i2 and

∫

T

∫

T

a 2 ( t ) dt are the signal energy

0

η2 ( t ) dt −

0

∫

T

ξ 2 ( t ) dt are the back-

0

ground noise. When considering the hypothesis H0 : Xi = ξi or X(t) = ξ(t) and the conditions ai = 0 or a(t) = 0 given ai* = ai or a(t) = a*(t), the left side of Equation (11.20) and Equation (11.21) has the following form:

∑

N i=1

ηi2 −

∑

N i=1

ξ i2 , or

∫

T

0

η2 ( t ) dt −

∫

T

ξ 2 ( t ) dt.

0

Subsequent analysis of the signal processing algorithm given by Equation (11.21) and Equation (11.22) will only be performed under the conditions ai* = ai or a(t) = a*(t). This statement is very important for further understanding of the generalized approach to signal processing in the presence of noise. How we do this becomes clear in References 50 and 51. It must be emphasized that

∑

N i=1

ηi2 −

∑

N i=1

ξ i2 → 0 as N → ∞ or

∫

T

0

η2 (t) dt −

∫

T

ξ 2 (t) dt → 0

0

as T → ∞ in the statistical sense because the processes ξi and ηi, or ξ(t) and η(t), are uncorrelated and have the same power spectral density of the additive Gaussian noise 0.5N0 according to the initial conditions. Thus, it has been shown that both the signal processing algorithms based on the observed input stochastic sample X1, …, XN and the two independently observed input stochastic samples X1, …, XN and η1, …, ηN have the same approach and are deﬁned by the likelihood function using the statistical theory of decision-making. The difference is that the numerator and denominator of the likelihood function used for the synthesis of the algorithms given by Equation (11.8) and Equation (11.9) involve the same observed input Copyright 2005 by CRC Press

1598_book.fm Page 433 Monday, October 11, 2004 1:57 PM

Foundations of the Generalized Approach to Signal Processing in Noise

433

stochastic sample [see Equation (11.4) and Equation (11.6)], but a “yes” signal is assumed in the numerator and a “no” signal in the denominator. The numerator of the likelihood function used for synthesis of the algorithm given by Equation (11.21) and Equation (11.22) involves the observed input stochastic sample, where a “yes” signal may be present and the denominator involves the reference sample, which is known a priori to contain a “no” signal. On this basis, it may be stated that only the sufﬁcient statistic

∫

∑

N i=1

Xi ai or

T

X ( t ) a ( t ) dt has been applied to deﬁne the mean of the likelihood function

0

in the signal processing algorithm given by Equation (11.8) and Equation (11.9), respectively. In the algorithm given by Equation (11.21) and Equation (11.22), the jointly sufﬁcient statistics dt and

∫

T

∑

N i=1

2 Xi ai and

∑

N i=1

( ηi2 − Xi2 ) or 2

∫

T

X(t ) a(t)

0

[η2 ( t ) − X 2 ( t )] dt are used to deﬁne the mean and variance of the

0

likelihood function. This fact permits us to obtain more complete information in the decision-making process in comparison to the algorithm given by Equation (11.8) and Equation (11.9). The algorithm given by Equation (11.21) and Equation (11.22) is free from a number of conditions unique to the algorithm given by Equation (11.8) and Equation (11.9). As the algorithm given by Equation (11.8) and Equation (11.9) is a component of the algorithm given by Equation (11.21) and Equation (11.22), the last has been called the generalized signal processing algorithm.

11.5 The Engineering Interpretation The technical realization of independent sampling from the regions Z and Z* obeying the same probability distribution density with the same statistical parameters is not difﬁcult. The solution of the problem of detecting the signal a(t) with the additive Gaussian noise n(t) is well known.1–49,55 The observed input stochastic process X(t) is examined at the output of the linear section of the receiver, which has an ideal amplitude-frequency response and the bandwidth ∆F. The noise at the linear section input of the receiver is the additive “white” Gaussian noise, having the correlation function 0.5N0δ(t2 – t1), where δ(x) is the delta function. The signal a(t) is assumed to be completely known, and the signal energy is taken to be equal to 1. The power spectral density 0.5N0 is considered an a priori indeterminate parameter. The gain of the linear section of the receiver

Copyright 2005 by CRC Press

1598_book.fm Page 434 Monday, October 11, 2004 1:57 PM

434

Signal and Image Processing in Navigational Systems

is equal to 1. Through analysis, the problem is reduced to testing the complex hypothesis with the decision function T

Re

∫

T

∫|X˙ (t)| dt ,

X˙ ( t ) a˙ ∗ ( t ) dt > K ( PF )

2

0

(11.25)

0

where a˙ * (t ) is the ﬁlter matched with the signal, PF is the threshold deﬁned T

by the probability of false alarm K(PF), and

∫ |X˙ (t)| dt is the statistic deﬁn2

0

ing the decision function. It turns out that the signal detector constructed in accordance with the above decision function renders the probability of false alarm PF stable, given an unknown noise power, and has the greatest probability of detection PD for any signal-to-noise ratio.28 Let us interpret this problem. We use two linear sections of the receiver (instead of one section) for our set of statistics. These linear sections will be called the preliminary (PF) and additional (AF) ﬁlters. The amplitude-frequency responses of PF and AF must obey the same law. The resonant frequencies of PF and AF must be detuned relative to each other by a value determined from the well-known results in References 56 and 57, for the purpose of providing uncorrelated statistics at the outputs of PF and AF. The detuning value between the resonant frequencies of PF and AF exceeds the effective signal bandwidth ∆Fa. As is well known,56,57 if this value reaches more than 4∆Fa, the coefﬁcient of correlation between the statistics at the outputs of PF and AF tends to approach zero. In practice, these statistics may be regarded as uncorrelated. The effective bandwidth of PF is equal to that of the signal frequency spectrum and can be even greater, but this is undesirable because the noise power at the output of PF is proportional to the effective bandwidth. The effective bandwidth of AF may be smaller than that of PF; however, for simplicity of analysis, in this chapter the effective bandwidth of AF is assumed to be the same as that of PF. Thus, we can assume that uncorrelated samples of the observed input stochastic process are formed at the outputs of PF and AF. These samples obey the same probability distribution density with the same statistical parameters given that the same process is present at the inputs of PF and AF, even if this process is the additive “white” Gaussian noise having the correlation function 0.5N0δ(t2 – t1). The physicotechnical interpretation of the signal processing algorithm given by Equation (11.21) and Equation (11.22) is the following. AF may serve as the source of the observed reference sample η1, …, ηN from the interference region Z*. The AF resonant frequency is detuned relative to the carrier frequency of the signal by a value that can be determined on the basis of well-known results,19,56,57 depending on the speciﬁc practical situation. PF serves as the source of the sample X1, …, XN of

Copyright 2005 by CRC Press

1598_book.fm Page 435 Monday, October 11, 2004 1:57 PM

Foundations of the Generalized Approach to Signal Processing in Noise

435

the observed input stochastic process from the interference region Z. The bandwidth of PF is matched with the effective bandwidth of the signal. The bandwidth value of PF is matched with that of AF. The ﬁrst term of the generalized signal processing algorithm, given by Equation (11.21) and Equation (11.22) corresponds to the synthesis of the correlation channel with twice the gain. The second term, given by Equation (11.21) and Equation (11.22), corresponds to the synthesis of the autocorrelation channel coupled with PF. The third term, given by Equation (11.21) and Equation (11.22), corresponds to the synthesis of the autocorrelation channel coupled with AF. The statistic of the autocorrelation channel coupled with PF is subtracted from the statistic of the autocorrelation channel coupled with AF. As a result,

∑

N i=1

ηi2 −

∑

N i=1

ξ i2 → 0 as N → ∞ or

∫

T

η2 ( t ) dt −

0

∫

T

ξ 2 ( t ) dt → 0 as N → ∞ in

0

the statistical sense. The statistic of the autocorrelation channel coupled with PF is subtracted from that of the correlation channel. As a result, a complete compensation of the noise component

∑

N i=1

ai* ξ i or

∫

T

a * (t )ξ(t ) dt of the sig-

0

nal processing algorithm given by Equation (11.8) and Equation (11.9) is achieved in the statistical sense if the conditions a*i = ai or a(t) = a*(t) are satisﬁed, where a*i or a*(t) is the model signal, and a1i , or a1 is the signal at the PF output. The detector shown in Figure 11.2 is based on the physicotechnical interpretation of the generalized approach to signal processing in noise50,51,58–61 stated in the preceding text.

xi =

{

ηi

AF

a1i + ξi ⇒ H 1

Yi =

ξi ⇒ H 0 PF

× ×

+ −

+

−

+ Z out = g

+

+ ×

−

{

Σ ΣNi=1(a21i + η2i − ξ2i) ⇒ H 1

ΣNi=1(η2i − ξ2i) ⇒ H 0

MSG – Model Signal Generator

FIGURE 11.2 The physicotechnical interpretation of the generalized detector.

Copyright 2005 by CRC Press

ni ⇒ H 0

+ MSG

×

{

ai + ni ⇒ H 1

1598_book.fm Page 436 Monday, October 11, 2004 1:57 PM

436

Signal and Image Processing in Navigational Systems

It is of special interest to compare these statements to the statements and analysis in Bacut et al.29 Some opponents of the generalized approach to signal processing in the presence of noise erroneously believe that this approach is the same as the one-input two-sample signal processing approach discussed in Bacut et al.29 For this purpose, we brieﬂy recall the main statements of the one-input two-sample signal processing algorithm. First, the signal sample is generated only at the time instants that correspond to the expected signal. In other words, the signal sample is generated only at the time instants when the expected signal may appear in the time frequency space. Second, the noise channel is formed within the limits of the time intervals in which it is a priori known that the signal is absent. The generalized approach to signal processing in the presence of noise is based on the following statements. The ﬁrst sample is generated independently of the time instants that correspond to the emergence of the signal in the time frequency space. The reference sample — the second sample — is formed simultaneously with the ﬁrst at the same time intervals as the ﬁrst and exists without any limitations in time or readings, but it is known a priori that a “no” signal exists in the reference sample due to the conditions for generating the reference sample. The sample sizes of the ﬁrst (signal) and second (reference) samples are the same. These differences between the generalized approach in the presence of noise and the one-input two-sample approach are very important. In addition, we can see that the engineering interpretation of the generalized approach in the presence of noise differs greatly from the one-input twosample approach in Bacut et al.29

11.6 Generalized Detector Let us consider the problem of speciﬁc interest, in which the signal has the stochastic amplitude and random initial phase. The necessity of considering this problem stems from the fact that, in practice, some satellite navigational systems use channels with an ionosphere mechanism of propagation and operate using frequencies that are higher than the maximal allowable frequency. Other satellite navigational systems use channels with tropospheric scattering. These problems arise in radar during detection of ﬂuctuating targets when the target return signal is a sequence of pulses of unknown amplitude and phase. A signal with the stochastic amplitude and random initial phase can be written in the following form: a ( t , A , ϕ 0 ) = A ( t ) S ( t ) cos[ω 0 t + Ψa ( t ) − ϕ 0 ],

Copyright 2005 by CRC Press

(11.26)

1598_book.fm Page 437 Monday, October 11, 2004 1:57 PM

Foundations of the Generalized Approach to Signal Processing in Noise

437

where ω0 is the carrier frequency of the signal a(t, A, ϕ0); S(t) is the known modulation law of the amplitude of the signal a(t, A, ϕ0); Ψa(t) is the known modulation law of the phase of the signal a(t, A, ϕ0); ϕ0 is the random initial phase of the signal a(t, A, ϕ0) which is uniformly distributed within the limits of the interval [–π, π] and is time variant within the limits of the time interval [0, T]; and A(t) is the amplitude factor, which is a random value and a function of time in the general case. Let us consider the generalized approach to signal processing in the presence of noise for the signals with the stochastic amplitude and random initial phase.63–66 According to the generalized approach, it is necessary to form the following: (1) the reference sample with a priori information of a “no” signal in the input stochastic process; (2) the autocorrelation channel for the purpose of compensating for the noise component of the correlation channel, which is caused by the interaction between the model signal and noise. The main principles of construction of the generalized detector for the signals with the stochastic amplitude and random initial phase using the Neyman–Pearson criterion are in References 66 to 70. The input stochastic process Y(t) must pass through the preliminary ﬁlter (PF). The effective bandwidth of PF is equal to ∆Fa , where ∆Fa is the effective spectrum bandwidth of the signal. X (t) = a 1 (t, A , ϕ 0 ) + ξ(t)

(11.27)

is the process at the output of PF, if a “yes” signal exists in the input stochastic process — the hypothesis H1. X (t) = ξ 1 (t)

(11.28)

is the process at the output of PF if a “no” signal exists in the input stochastic process — the hypothesis H0 ; a1(t, A, ϕ0) is the signal at the output of PF given by Equation (11.26); and ξ(t) is the noise at the output of PF. We need to form the reference sample for the generation of the jointly sufﬁcient statistics of the likelihood function mean and variance. For this purpose, the additional ﬁlter (AF) is formed parallel to PF. The amplitudefrequency response of AF is analogous over the entire range of parameters to that of PF, but it is detuned in the resonant frequency relative to PF for the purpose of providing uncorrelated statistics at the outputs of PF and AF. The detuning value must be larger than the effective spectrum bandwidth of the signal so that the processes at the outputs of PF and AF will be uncorrelated. As was shown in References 56 and 57, if this detuning value reaches the value between 4∆Fa and 5∆Fa , the processes at the outputs of PF and AF are not correlated practically. In this condition, the correlation coefﬁcient between the statistics at the outputs of PF and AF is not more than 0.05 for all practical purposes. It may be considered as a value tending to approach zero. Thus, the process η(t) is formed at the output of AF: Copyright 2005 by CRC Press

1598_book.fm Page 438 Monday, October 11, 2004 1:57 PM

438

Signal and Image Processing in Navigational Systems η( t ) = ξ 2 ( t ) cos[ω ′n t + υ′ ( t )] ,

(11.29)

where ξ2(t) is the random envelope of the amplitude of the noise at the output of AF; υ′(t) is the random phase of the noise at the output of AF; and ωn′ is the medium frequency of the noise at the output of AF. The generalized detector for the signals with the stochastic amplitude and random initial phase is shown in Figure 11.3. There are two ﬁlters with the nonoverlapping amplitude-frequency responses that must obey the same law: PF and the AF. PF is matched with the effective spectrum bandwidth of the signal with the carrier frequency ω0 . AF does not pass the frequency ω0 . By this means, there is the following requirement for AF: the resonant frequency of AF must be detuned with respect to that of PF to ensure the uncorrelated statistics at the outputs of both AF and PF.66–72 This requirement is necessary to ensure the complete compensation of the background noise constant component at the output of the generalized detector in the statistical sense. For the generation of the jointly sufﬁcient statistics of the likelihood function mean and variance, in accordance with the generalized approach to signal processing in the presence of noise, it is necessary to form the autocorrelation channel. These actions allow us to compensate for the total noise component in the statistical sense. We now proceed to show this. The generalized detector for the signals with the stochastic amplitude and random initial phase shown in Figure 11.3 consists of the correlation channel (the multipliers 1 and 2, the integrators 1 and 2, the square-law function generators 5 and 6, the summator 2, the model signal generator MSG); the autocorrelation channel [the multipliers 3 and 4, the summator 1, the integrator 3, the square-law function generator 7, the ampliﬁer (>)]; the compensating channel (the summators 3 and 4, the compensation of the total noise component in the statistical sense is carried out by the summators 3 and 4); the delay blocks 1–5 are only used for speciﬁc technical problems, and are not taken into consideration during the analysis of the theoretical principles of functionality. The compensating channel of the generalized detector allows us to compensate the correlation channel noise component and the autocorrelation channel random component both of the generalized detector, in the statistical sense. The noise component is created by the interaction between the model signal and noise. The random component, which will be described below, is caused by the interaction between the signal and noise. Let us analyze the generalized detector shown in Figure 11.3 for the condition S(t) = S*(t),

(11.30)

i.e., the model signal a*(t) is completely matched with the signal a1(t, A, ϕ0) at the output of PF, and τ = 0. How we are able to do this becomes clear in the discussion of the experimental and application results presented in References 50 and 51. We must take into consideration that the frequencies 2ω0, 2ωn, or 2ωn′ cannot pass through PF and AF, respectively. The analysis is

Copyright 2005 by CRC Press

1598_book.fm Page 439 Monday, October 11, 2004 1:57 PM

Foundations of the Generalized Approach to Signal Processing in Noise

x(t ) =

{

η(t ) a1(t ) + ξ(t ) ξ(t ) ⇒ H 0

PF

delay τ

0

Sm(t ) cos[ω 0t + Ψa (t )] Sm(t ) sin[ω 0t + Ψa (t )]

3

+

1

+

6 5

0

3

+

√ Z g2(t )

−

−

+

3

+

>

×

∫

− 4

delay τ

T

4

+

×

delay τ

+

×

+

4

3

×

⇒ H0

2

2

1

delay τ

n(t )

+

delay τ

T

∫

0

⇒ H1

×

MSG

2

a(t ) + n (t )

5

T

∫

×

{

Y(t ) =

1

1

×

AF

⇒ H1

439

7

MSG – Model Signal Generator 2

FIGURE 11.3 The generalized detector for the signal with stochastic parameters.

based on the results discussed in References 62 to 678 and the hypothesis H1 — a “yes” signal in the input stochastic process. In terms of Equation (11.26), the processes at the outputs of the multipliers 1–4 take the following form: y1 (t) = [ a1 (t , A, ϕ 0 ) + ξ(t)]a1* (t) = { A(t)S(t) cos[ω 0 t + Ψ a (t) − ϕ 0 ] + ξ1 (t) cos[ω nt + υ(t)]}S* (t) cos[ω 0 t + Ψ a (t)] = 0.5 cos ϕ 0 A(t)S(t)S* (t) + 0.5S* (t)ξ1 (t) cos[Ψ a (t) − υ(t)] ; (11.31) y2 (t) = [ a1 (t , A, ϕ 0 ) + ξ(t)]a2* (t) = { A(t)S(t) cos[ω 0 t + Ψ a (t) − ϕ 0 ] + ξ1 (t) cos[ω nt + υ(t)]}S* (t)sin[ω 0 t + Ψ a (t)] = 0.5 sin ϕ 0 A(t)S(t)S* (t) + 0.5S* (t)ξ1 (t)sin[Ψ a (t) − υ(t)] ; (11.32) Copyright 2005 by CRC Press

1598_book.fm Page 440 Monday, October 11, 2004 1:57 PM

440

Signal and Image Processing in Navigational Systems y3 (t) = 0.5 A2 (t)S2 (t) + A(t)S(t)ξ1 (t) cos[Ψ a (t) − υ(t) − ϕ 0 ] + 0.5 ξ12 (t) ; (11.33) y4 (t) = η2 (t) = ξ 2 (t) cos[ω ′n + υ′(t)] ξ 2 (t) cos[ω ′n + υ′(t)] = 0.5 ξ 22 (t) , (11.34)

where a1* (t ) = S* (t ) cos[ω 0 t + Ψa (t )] and a2* (t ) = S* (t ) sin[ω 0 t + Ψa (t )] (11.35) is the model signal, which is formed at the output of the model signal generator MSG (Figure 11.3). The notation S*(t) is retained to emphasize the interaction between the model signal and noise. The process at the output of summator 1 takes the following form: y 5 ( t ) = 0.5A 2 ( t ) S 2 ( t ) + A ( t ) S ( t )ξ 1 ( t ) cos[Ψa ( t ) − υ( t ) − ϕ 0 ] − 0.5[ξ 22 ( t ) − ξ 21 ( t )] . (11.36) The processes at the outputs of integrators 1–3 take the following form: T

T

∫

∫

Z1 (t ) = 0.5 cos ϕ0 A(t )S(t )S (t ) dt + 0.5 S* (t )ξ1 (t ) cos[Ψa (t ) − υ(t )] dt ; *

0

0

(11.37) T

T

∫

∫

Z2 (t ) = 0.5 sin ϕ0 A(t )S(t )S (t ) dt + 0.5 S* (t )ξ1 (t ) sin[Ψa (t ) − υ(t )] dt ; *

0

0

(11.38) T

∫

Z3 (t) = 0.5 A2 (t)S2 (t) dt 0

T

T

+

∫ A(t)S(t)ξ (t) cos[Ψ (t) − υ(t) − ϕ ] dt − 0.5 ∫ [ξ (t) − ξ (t)] dt . 1

0

a

2 2

0

2 1

0

(11.39) It should be particularly emphasized that all integration operations must be read in the statistical sense. Let us consider the process Z3(t) at the output of integrator 3 given by Equation (11.39). The ﬁrst term of the process Z3(t) is proportional to the energy of the signal within the limits of the time interval [0, T]. The second term is the random component of the autocorrelation channel of the Copyright 2005 by CRC Press

1598_book.fm Page 441 Monday, October 11, 2004 1:57 PM

Foundations of the Generalized Approach to Signal Processing in Noise

441

generalized detector, which is caused by the interaction between the signal and noise. The third term is the difference between the powers of the noise, which are formed at the outputs of PF and AF, respectively. The third term does not participate in compensating the noise components of the generalized detector correlation channel

∫

T

S(t) S*(t) dt

∫

T

0

0

A(t )S* (t)ξ1(t) cos[Ψa(t) –

2

 T  υ(t) – ϕ0]dt and  S* (t )ξ1 (t ) dt  , which are caused by the interaction  0  between the model signal and noise. The third term has the following phys-

∫

ical sense. The integral

∫

T

0

[ξ 22 ( t ) − ξ 21 (t)] dt is the background noise at the

output of the generalized detector and tends to approach zero as T → ∞ in the statistical sense (see Section 11.4). The background noise is only used for deﬁnition of the threshold Kg during decision making. Based on this statement, the third term of the process Z3(t) given by Equation (11.39) can be discarded in the following analysis, but we will take it into account in the end result. The processes at the outputs of the square-law function generators 1 and 2 take the following form: T

T

∫

∫

Z12 (t) = 0.25 cos 2 ϕ 0 A2 (t)S(t)S∗ (t) dt S(t)S* (t) dt 0

0

T    + 0.25  S* (t)ξ1 (t) cos[Ψ a (t) − υ(t)] dt   0 

2

∫

(11.40)

T

T

∫

∫

+ 0.5 cos ϕ 0 S(t)S (t) dt A(t)S* (t)ξ1 (t) cos[Ψ a (t) − υ(t)] dt , *

0

0

T

∫

T

∫

Z (t) = 0.25 sin ϕ 0 A (t)S(t)S (t) dt S(t)S* (t) dt 2 2

2

2

*

0

0

T    + 0.25  S* (t)ξ1 (t)sin[Ψ a (t) − υ(t)] dt   0 

2

∫

T

∫

T

∫

+ 0.5 sin ϕ 0 S(t)S* (t) dt A(t)S* (t)ξ1 (t)sin[Ψ a (t) − υ(t)] dt . 0

Copyright 2005 by CRC Press

0

(11.41)

1598_book.fm Page 442 Monday, October 11, 2004 1:57 PM

442

Signal and Image Processing in Navigational Systems

Using the straightforward mathematical transformations in References 25 and 33, it is not difﬁcult to show that  T   0.25  S* (t)ξ1 (t) cos[Ψ a (t) − υ(t)] dt   0 

2

∫

T    = 0.25  S* (t)ξ1 (t)sin[Ψ a (t) − υ(t)] dt   0 

2

∫

(11.42)

2

 T   = 0.25  S* (t)ξ1 (t) dt  .   0

∫

The process at the output of summator 2 in terms of Equation (11.42) has the following form:  T  Z (t ) = 0.25 A (t )S(t )S (t ) dt S(t )S (t ) dt + 0.25 S* (t )ξ1 (t ) dt   0  0 0 T

∫

2 Σ

T

2

*

T

∫

2

∫

*

T

∫

∫

+ 0.5 S(t )S* (t ) dt A(t )S* (t )ξ1 (t ) cos[Ψa (t ) − υ(t ) − ϕ0 ] dt . 0

0

(11.43) The process at the output of the square-law function generator 7 in terms of Equation (11.42) takes the following form: T

T

∫

∫

Z32 (t) = 0.25 A2 (t)S(t)2 dt A2 (t)S2 (t) dt 0

0

T    + 0.25  A(t)S∗ (t)ξ1 (t) dt   0 

2

∫

T

+

(11.44)

T

∫ A (t)S (t) dt ∫ A(t)S(t)ξ (t) cos[Ψ (t) − υ(t) − ϕ ] dt. 2

2

1

0

a

0

0

Considering the second and third terms in Equation (11.43) and Equation (11.44), respectively, we can see that they differ by the factor A2(t) under the condition given by Equation (11.30), and the second and third terms in Equation (11.42) and Equation (11.43) agree within the factor of 2. Copyright 2005 by CRC Press

1598_book.fm Page 443 Monday, October 11, 2004 1:57 PM

Foundations of the Generalized Approach to Signal Processing in Noise

443

Before proceeding to questions of compensation of these terms in the statistical sense, let us consider two plausible cases. The ﬁrst case implies that the amplitude envelope of the signal is not the stochastic function of time within the limits of the time interval [0, T] for a single sample and can be stochastic from sample to sample — the case of the slow ﬂuctuations. This case is very important for certain radar navigational systems where the signal may occur over and over. The second case is based on the fact that the envelope of the amplitude of the signal is the stochastic function of time within the limits of the time interval [0, T] for a single sample, or in other words, the case of the rapid ﬂuctuations of the envelope of amplitude of the signal. Let us consider both cases in detail. 11.6.1 The Case of the Slow Fluctuations In this case, the compensation between the second and third terms in Equation (11.43) and Equation (11.44) in the statistical sense is conceivable by averaging on a set M of realizations of the input stochastic process X(t). All statistical characteristics of the input stochastic process X(t) are invariant within the limits of the time interval [0, T]. Then, Equation (11.44) may be written in the following form: M

M

T T    2 2 2 2  Aj (t)Sj (t) dt Aj (t)Sj (t) dt   0  0

∑ Z (t) = 0.25∑ ∫ 2 3j

j=1

j=1

∫

 T    Aj (t)Sj (t)ξ1j (t) dt    0

M

∑∫

+ 0.5

j=1

T

M

+

2

T

∑{∫ A (t)S (t) dt ∫ A (t)S (t)ξ 2 j

j=1

2 j

j

j

1j

}

(t) cos[Ψ aj (t) − υ j (t) − ϕ 0j ] dt .

0

0

(11.45) Equation (11.43) may be written in identical form: M

∑Z

2 Σj

M

T T    2 * *  Aj (t)Sj (t)Sj dt Sj (t)Sj (t) dt  0  0 

∑∫

(t) = 0.25

j=1

j=1

M

∫

T   *   Sj (t)ξ1j (t) dt   0 

∑∫

+ 0.25

j=1

M

T T    * * ( ) ( ) ( t ) S ( t ) ξ ( t ) cos[ Ψ ( t ) − υ ( t ) − ϕ ] dt S t S t dt A  j . j 1j aj j 0j j j  0  0 (11.46)

∑∫

+ 0.5

j=1

Copyright 2005 by CRC Press

2

∫

1598_book.fm Page 444 Monday, October 11, 2004 1:57 PM

444

Signal and Image Processing in Navigational Systems

We introduce the designations for the condition given by Equation (11.30) T

E a1 j =

T

∫ S (t)S (t) dt ∗ j

j

∫

σ E a1 j = A 2j ( t ) S j ( t ) S j∗ ( t ) dt , (11.47) 2 Aj

and

0

0

where Ea1j is the energy of the signal within the limits of the time interval in the j-th realization of the input stochastic process, and σA2j is the variance of the amplitude envelope factor Aj(t) of the signal within the limits of the time interval [0, T] in the j-th realization of the input stochastic process. An analogous representation of the variance σA2j is used in References 8 and 16.8,16 Let us consider the second term in Equation (11.45). This term can be represented as a product of two integrals:   j=1  

∑ ∫ M

T

0

2

 Aj (t)Sj (t)ξ1j (t) dt  = 



∑  ∫ dt ∫ M

j=1

T

T

0

0

 Aj2 (t)Sj (t)Sj (τ)ξ1j (t)ξ1j (τ) dτ . 

Averaging the integrand with respect to the amplitude envelope factor Aj(t) in the j-th realization for a set M of realizations of the input stochastic process, we can write T

T

T

∫ ∫

dt A 2j ( t )S j ( t ) S j (τ )ξ 1j ( t )ξ 1j (τ ) dτ = σ 2Aj

0

0

T

∫ ∫

dt S j ( t ) S j (τ )ξ 1j ( t )ξ 1j (τ ) dτ .

0

0

(11.48) In terms of Equation (11.47) and Equation (11.48), Equation (11.45) and Equation (11.46) can take the following form:

∑ Z (t) = 0.25∑ σ 2 3j

j=1

∑

+ 0.5

j=1

Ea1j

∑ j=1

∫ A (t)S (t) dt 2 j

2 j

0

 T   σ  Sj (t)ξ1j (t) dt   0 

2

∫

2 Aj

T

M

+

2 Aj

j=1

M

Copyright 2005 by CRC Press

T

M

M

σ 2Aj Ea1j

∫ A (t)S (t)ξ j

0

j

1j

(t) cos[Ψ aj (t) − υ j (t) − ϕ 0j ] dt ;

(11.49)

1598_book.fm Page 445 Monday, October 11, 2004 1:57 PM

Foundations of the Generalized Approach to Signal Processing in Noise

∑

T

M

M

445

∑ ∫ A (t)S (t)S (t) dt

ZΣ2 j (t) = 0.25

j=1

j=1

M

∑

+ 0.25

j=1

2 j

Ea1j

*

j

0

T   2  σ Aj  Sj* (t)ξ1j (t) dt   0 

2

∫

(11.50)

T

M

∑ E ∫ A (t)S (t)ξ

+ 0.5

* j

j

a1 j

j=1

1j

(t) cos[Ψ aj (t) − υ j (t) − ϕ 0j ] dt.

0

Comparing Equation (11.49) with Equation (11.50), we can see that the second and third terms differ by value

∑

M j=1

σ 2Aj and by the factor 2. Because of this,

the ampliﬁer (>) of the autocorrelation channel of the generalized detector has the ampliﬁcation factor

1 σ 2A

and is connected to the input of the compen-

sating channel of the generalized detector. Taking this fact into account, we may write Equation (11.49) in the following form: T

M

M

∑ Z (t) = 0.25∑ σ 2 3j

j=1

Ea1j

j=1

M

∑

+ 0.5

j=1

∑σ

∫ A (t)S (t) dt 2 j

2 j

0

T    σ  Sj (t)ξ1j (t) dt   0 

2

∫

2 Aj

(11.51)

T

M

+

2 Aj

2 Aj

Ea1j

j=1

∫ A (t)S (t)ξ j

j

1j

(t) cos[Ψ aj (t) − υ j (t) − ϕ 0j ] dt .

0

Taking into consideration the condition given by Equation (11.30) and the discarded third term of the process Z3(t) in Equation (11.39), the process at the output of the compensating channel of the generalized detector — the output of the summator 4 — takes the following form: M

[Zgout (t)] 2 = 0.25∑ Ea

1j

j=1

T

M

∫ A (t)S (t) dt + 0.25∑ ∫ 2 j

0

2 j

j=1

2

 T   2 2  [ξ 2j (t) − ξ1j (t)] dt  .   0 (11.52)

Reference to Equation (11.52) shows that the compensation between the second and third terms in Equation (11.43) and Equation (11.44), in the statistical sense, is performed at the output of the compensating channel of

Copyright 2005 by CRC Press

1598_book.fm Page 446 Monday, October 11, 2004 1:57 PM

446

Signal and Image Processing in Navigational Systems

the generalized detector — the output of summator 4 — during averaging on a set M of realizations of the input stochastic process X(t). It should be considered that all integration operations are implied in the statistical sense. In this conjunction, the process at the output of the generalized detector takes the following form: T

M

M

2

T    2 2 [ ξ ( t ) − ξ ( t )] dt   . (11.53) 2j 1j  0 

∑ E ∫ A (t)S (t) dt + ∑ ∫

Z (t) = 0.5 out g

2 j

a1 j

j=1

2 j

j=1

0

The ﬁrst term in Equation (11.53), at the output of the generalized detector, is the energy of the signal, and the second term is the background noise that tends to approach zero in the statistical sense. 11.6.2 The Case of the Rapid Fluctuations In this case, the envelope of the amplitude of the signal is the stochastic function of the time within the limits of the time interval [0, T] for a single realization of the input stochastic process X(t). The process at the output of summator 2 — the output of the correlation channel of the generalized detector — takes the following form: T

T

∫

∫

Z (t) = 0.25 S (t)S(t) dt A2 (t)S* (t)S(t) dt 2 Σ

*

0

0

T    . + 0 25  S* (t)ξ1 (t) dt   0 

2

∫

T

(11.54)

T

∫

∫

+ 0.5 S (t)S(t) dt A(t)S* (t)ξ1 (t)cvos[Ψ a (t) − υ(t) − ϕ 0 ] dt. *

0

0

The process at the output of the autocorrelation channel of the generalized detector — the output of square-law function generator 7 — takes the following form: T

T

∫

∫

Z32 (t) = 0.25 A2 (t)S2 (t) dt A2 (t)S2 (t) dt 0

0

T    + 0.5  A(t)S(t)ξ1 (t) dt   0 

2

∫

T

+

T

∫ A (t)S (t) dt ∫ A(t)S(t)ξ (t) cos[Ψ (t) − υ(t) − ϕ ] dt. 2

2

1

0

Copyright 2005 by CRC Press

(11.55)

0

a

0

1598_book.fm Page 447 Monday, October 11, 2004 1:57 PM

Foundations of the Generalized Approach to Signal Processing in Noise

447

We proceed to introduce the designations in accordance with the results discussed in References 71 to 74 for the condition given by Equation (11.30): T

∫

T

∫

E a1 = S (t )S(t ) dt and σ E a1 = A2 (t )S* (t )S(t ) dt , *

2 A

0

(11.56)

0

where Ea1 is the energy of the signal within the limits of the time interval [0, T], and σA2 is the variance of the amplitude envelope factor A(t) within the limits of the time interval [0, T]. Let us consider the second term in Equation (11.55) that can be expressed as the product of two integrals.25 Let us consider the double integral 2

T T  T   A(t )S(t )ξ1 (t ) dt  = dt A(t ) A(τ)S(t )S(τ)ξ1 (t )ξ1 (τ) dτ. Averaging the inte 0  0 0 grand with respect to the amplitude envelope factor A(t) within the limits of the time interval [0, T], we can write

∫

∫ ∫

T

T

T

T

0

0

∫ dt ∫ A(t)A(τ)S(t)S(τ)ξ (t)ξ (τ) dτ = σ ∫ dt ∫ S(t)S(τ)ξ (t)ξ (τ) dτ . 1

0

2 A

1

0

1

1

(11.57) Equation (11.54) and Equation (11.55) take the following form in terms of Equation (11.56) and Equation (11.57): T

Z (t) = 0.25Ea1 2 Σ

∫ 0

T    A (t)S (t)S(t) dt + 0.25  S* (t)ξ1 (t) dt   0  2

2

∫

*

(11.58)

T

+ 0.5Ea1

∫ A(t)S (t)ξ (t) cos[Ψ (t) − υ(t) − ϕ ]dt; *

a

1

0

0

T

Z (t) = 0.25σ E 2 3

2 A a1

∫ 0

 T   A (t)S (t) dt + 0.5σ  S(t)ξ1 (t) dt    0 2

2

2 A

∫

T

+ σ 2AEa1

∫ A(t)S(t)ξ (t) cos[Ψ (t) − υ(t) − ϕ ]dt . 1

0

Copyright 2005 by CRC Press

a

2

0

(11.59)

1598_book.fm Page 448 Monday, October 11, 2004 1:57 PM

448

Signal and Image Processing in Navigational Systems

Comparing Equation (11.58) with Equation (11.59), we can see that the second and third terms differ by the variance σA2 of the amplitude envelope factor and a factor of 2. Because of this, the ampliﬁer (>) of the generalized detector autocorrelation channel has the ampliﬁcation factor

1 σ 2A

and is connected to

the input of the generalized detector compensating channel. Taking this fact into consideration, Equation (11.59) can be written in the following form: T

Z (t) = 0.25Ea1 2 3

∫ 0

T    A (t)S (t) dt + 0.5  S(t)ξ1 (t) dt   0  2

2

∫

2

(11.60)

T

+ Ea1

∫ A(t)S(t)ξ (t) cos[Ψ (t) − υ(t) − ϕ ] dt . 1

a

0

0

For the condition given by Equation (11.30) and in terms of the discarded third term of the process Z3(t) in Equation (11.39), the process at the output of the generalized detector compensating channel — the output of summator 4 — takes the following form: T

[Zgout (t)] 2 = 2ZΣ2 (t) − Z32 (t) = 0.25Ea

1

∫ A (t)S (t) dt 2

2

0

(11.61)

2

T    + 0.25  [ξ 22 (t) − ξ12 (t)] dt  .   0

∫

Equation (11.61) shows that the compensation between the second and third terms in Equation (11.43) and Equation (11.44), in the statistical sense, is performed at the output of the generalized detector compensating channel — the output of summator 4 — during averaging of the input stochastic process X(t) within the limits of the time interval [0, T]. It should be noted that all integration operations are implied in the statistical sense. The process at the output of the generalized detector takes the following form: T

Z

out g

∫

T

( t ) = 0.5 E a1 A ( t ) S ( t ) dt + 0

2

2

{∫ [ξ (t) − ξ (t)] dt} 2 2

2 1

2

.

(11.62)

0

The ﬁrst term in Equation (11.62) is the energy of the signal at the output of the generalized detector, and the second term is the background noise at the

Copyright 2005 by CRC Press

1598_book.fm Page 449 Monday, October 11, 2004 1:57 PM

Foundations of the Generalized Approach to Signal Processing in Noise

449

output of the generalized detector that tends to approach zero in the statistical sense. Thus, if the envelope of amplitude of the signal is the stochastic function of the time within the limits of the time interval [0, T], it is possible to perform the compensation between the correlation channel noise component of the generalized detector [which is created by the interaction between the model signal and noise — the second and third terms in Equation (11.54)] and the random component of the autocorrelation channel of the generalized detector [which is caused by the interaction between the signal and noise — the second and third terms in Equation (11.55)], without averaging of the input stochastic process X(t) on a set of realizations. This effect of compensation takes place within the limits of the time interval [0, T]. To attain these ends, the ampliﬁcation factor of the generalized detector’s autocorrelation channel ampliﬁer (>) must differ by

1 σ 2A

in comparison with the correlation channel

ampliﬁcation factor. Generation of the jointly sufﬁcient statistics of the likelihood function mean and variance during the use of the generalized detector for the signals with the stochastic amplitude and random initial phase allows us, in principle, to compensate the generalized detector’s correlation channel noise component (which is created by the interaction between the model signal and noise) and its autocorrelation channel random component (which is created by the interaction between the signal and noise), using its compensating channel. The generalized detector allows us to increase the signal-to-noise ratio at the output of the detector in comparison to the optimal detectors of the classical and modern signal processing theories during the same input conditions.

11.7 Conclusions Let us summarize brieﬂy the main results discussed in this chapter. The proposed modiﬁcation of the initial premises of the classical and modern signal processing theories assumes that the frequency–time regions of the noise exist where a “yes” signal may be found, and where it is known a priori that a “no” signal exists. This modiﬁcation allows us to perform the theoretical synthesis of the generalized signal processing algorithm. Two uncorrelated samples are used, one of which is the reference sample because it is known a priori that a “no” signal is found in this sample. This fact allows us to obtain the jointly sufﬁcient statistics of the likelihood function mean and variance. The optimal signal processing algorithms of the classical and modern theories for the signals with known and unknown amplitude-phasefrequency structure allow us to obtain only the sufﬁcient statistic of the likelihood function mean and are components of the generalized signal processing algorithm.

Copyright 2005 by CRC Press

1598_book.fm Page 450 Monday, October 11, 2004 1:57 PM

450

Signal and Image Processing in Navigational Systems

The physicotechnical interpretation of the generalized approach to signal processing in the presence of noise is a combination of the optimal approaches of the classical and modern theories for signals with both known and unknown amplitude-phase-frequency structure. The additional ﬁlter (AF) is the source of the reference sample. The resonant frequency of AF is detuned relative to that of the preliminary ﬁlter (PF). The value of the detuning is greater than the effective spectral bandwidth of the signal. The use of AF jointly with PF forms the background noise at the output of the generalized detector. The background noise is the difference between the energy characteristics of the noise at the outputs of PF and AF and tends to approach zero in the statistical sense. In other words, the background noise at the output of the generalized detector is formed as a result of the generation of the jointly sufﬁcient statistics of the likelihood function mean and variance for the generalized approach to signal processing in the presence of noise. The background noise is caused by both the noise at the output of the PF and the noise at the output of the AF. The background noise at the output of the generalized detector is independent of both the signal and the model signal. The correlation between the noise component

∑

N i=1

ai∗ξ i or

∫

T

a ∗ ( t )ξ ( t ) dt

0

of the generalized detector’s correlation channel and the autocorrelation channel random component

∑

N i=1

a1i ξ i or

∫

T

0

a 1 ( t )ξ ( t ) dt allows us to gener-

ate the jointly sufﬁcient statistics of the likelihood function mean and variance. The noise component of the correlation channel is caused by the interaction between the model signal and noise. The random component of the autocorrelation channel is caused by the interaction between the signal and noise. The effect of compensation between these two components is caused by the generation of the jointly sufﬁcient statistics of the likelihood function mean and variance for the generalized approach to signal processing in the presence of noise and employing generalized detectors in various complex navigational systems. The effect of this compensation is carried out within the limits of the sample size [1, N] or of the time interval [0, T], for which the input stochastic process is observed. The use of the generalized detector for the signals with the stochastic amplitude and random initial phase in various complex navigational systems has the following peculiarity. If the envelope of the amplitude of the signal is not a stochastic function of time within the limits of the time interval [0, T] for a single realization of the input stochastic process and can be stochastic from realization to realization, the generation of the jointly sufﬁcient statistics of the likelihood function mean and variance is carried out by averaging on a set M of realizations of the input stochastic process. If the envelope of the amplitude of the signal is a stochastic function of time within the limits of the time interval [0, T], then the generation of the jointly sufﬁcient statistics of the likelihood function mean and variance is possible within Copyright 2005 by CRC Press

1598_book.fm Page 451 Monday, October 11, 2004 1:57 PM

Foundations of the Generalized Approach to Signal Processing in Noise

451

the limits of that time interval without averaging the input stochastic processes on a set of realizations.

References 1. Kotelnikov, V., Potential Noise Immunity Theory, Soviet Radio, Moscow, 1956 (in Russian). 2. Wiener, N., Non-Linear Problems in Stochastic Process Theory, McGraw-Hill, New York, 1959. 3. Middleton, D., An Introduction to Statistical Communication Theory, McGrawHill, New York, 1961. 4. Shannon, K., Research on Information Theory and Cybernetics, McGraw-Hill, New York, 1961. 5. Wiener, N., Cybernetics or Control and Communication in the Animal and the Machine, 2nd ed., John Wiley & Sons, New York, 1961. 6. Selin, I., Detection Theory, Princeton University Press, Princeton, NJ, 1965. 7. Miller, R., Simultaneous Statistical Inference, McGraw-Hill, New York, 1966. 8. Van Trees, H., Detection, Estimation, and Modulation Theory. Part 1: Detection, Estimation, and Linear Modulation Theory, John Wiley & Sons, New York, 1968. 9. Helstrom, C., Statistical Theory of Signal Detection, 2nd ed., Pergamon Press, Oxford, 1968. 10. Gallager, R., Information Theory and Reliable Communication, John Wiley & Sons, New York, 1968. 11. Thomas, J., An Introduction to Statistical Communication Theory, John Wiley & Sons, New York, 1969. 12. Jazwinski, A., Stochastic Processes and Filtering Theory, Academic Press, New York, 1970. 13. Van Trees, H., Detection, Estimation, and Modulation Theory. Part 2: Non-Linear Modulation Theory, John Wiley & Sons, New York, 1970. 14. Schwartz, M., Information, Transmission, Modulation, and Noise, 2nd ed., McGraw-Hill, New York, 1970. 15. Wong, E., Stochastic Processes in Information and Dynamical Systems, McGrawHill, New York, 1971. 16. Van Trees, H., Detection, Estimation, and Modulation Theory. Part 3: Radar-Sonar Signal Processing and Gaussian Signals in Noise, John Wiley & Sons, New York, 1972. 17. Box, G. and Tiao, G., Byesian Inference in Statistical Analysis, Addison-Wesley, Cambridge, MA, 1973. 18. Stratonovich, R., Principles of Adaptive Processing, Soviet Radio, Moscow, 1973 (in Russian). 19. Levin, B., Theoretical Foundations of Statistical Radio Engineering, Parts 1–3, Soviet Radio, Moscow, 1974–1976 (in Russian). 20. Tikhonov, V. and Kulman, N., Non-Linear Filtering and Quasideterministic Signal Processing, Soviet Radio, Moscow, 1975 (in Russian). 21. Repin, V. and Tartakovsky, G., Statistical Synthesis under a priori Uncertainty and Adaptation of Information Systems, Soviet Radio, Moscow, 1977 (in Russian).

Copyright 2005 by CRC Press

1598_book.fm Page 452 Monday, October 11, 2004 1:57 PM

452

Signal and Image Processing in Navigational Systems

22. Kulikov, E. and Trifonov, A., Estimation of Signal Parameters in Noise, Soviet Radio, Moscow, 1978 (in Russian). 23. Sosulin, Yu, Detection and Estimation Theory of Stochastic Signals, Soviet Radio, Moscow, 1978 (in Russian). 24. Ibragimov, I. and Rozanov, Y., Gaussian Random Processes, Springer-Verlag, New York, 1978. 25. Anderson, B. and Moore, J., Optimal Filtering, Prentice Hall, Englewood Cliffs, NJ, 1979. 26. Shirman, Y. and Manjos, V., Theory and Methods in Radar Signal Processing, Radio and Svyaz, Moscow, 1981 (in Russian). 27. Huber, P., Robust Statistics, John Wiley & Sons, New York, 1981. 28. Blachman, N., Noise and Its Effect in Communications, 2nd ed., Krieger, Malabar, FL, 1982. 29. Bacut, P. et al., Signal Detection Theory, Radio and Svayz, Moscow, 1984 (in Russian). 30. Anderson, T., An Introduction to Multivariate Statistical Analysis, 2nd ed., John Wiley & Sons, New York, 1984. 31. Lehmann, E., Testing Statistical Hypothesis, 2nd ed., John Wiley & Sons, New York, 1986. 32. Silverman, B., Density Estimation for Statistics and Data Analysis, Chapman & Hall, London, 1986. 33. Bassevillee, M. and Benveniste, A., Detection of Abrupt Changes in Signals and Dynamical Systems, Springer-Verlag, Berlin, 1986. 34. Trifonov, A. and Shinakov, Yu, Joint Signal Differentiation and Estimation of Signal Parameters in Noise, Radio and Svayz, Moscow, 1986 (in Russian). 35. Thomas, A., Adaptive Signal Processing: Theory and Applications, John Wiley & Sons, New York, 1986. 36. Blahut, R., Principles of Information Theory, Addison-Wesley, Reading, MA, 1987. 37. Weber, C., Elements of Detection and Signal Design, Springer-Verlag, New York, 1987. 38. Skolnik, M., Radar Applications, IEEE Press, New York, 1988. 39. Kassam, S., Signal Detection in Non-Gaussian Noise, Springer-Verlag, Berlin, 1988. 40. Poor, V., Introduction to Signal Detection and Estimation, Springer-Verlag, New York, 1988. 41. Brook, D. and Wynne, R., Signal Processing: Principles and Applications, Pentech Press, London, 1988. 42. Porter, W. and Kak, S., Advances in Communications and Signal Processing, Springer-Verlag, Berlin, 1989. 43. Adrian, C., Adaptive Detectors for Digital Modems, Pentech Press, London, 1989. 44. Scharf, L., Statistical Signal Processing, Detection, Estimation, and Time Series Analysis, Addison-Wesley, Reading, MA, 1991. 45. Cover, T. and Thomas, J., Elements of Information Theory, John Wiley & Sons, New York, 1991. 46. Basseville, M. and Nikiforov, I., Detection of Abrupt Changes, Prentice Hall, Englewood Cliffs, NJ, 1993. 47. Dudgeon, D. and Johnson, D., Array Signal Processing: Concepts and Techniques, Prentice Hall, Englewood Cliffs, NJ, 1994. 48. Porat, B., Digital Processing of Random Signals: Theory and Methods, Prentice Hall, Englewood Cliffs, NJ, 1994.

Copyright 2005 by CRC Press

1598_book.fm Page 453 Monday, October 11, 2004 1:57 PM

Foundations of the Generalized Approach to Signal Processing in Noise

453

49. Helstrom, C., Elements of Signal Detection and Estimation, Prentice Hall, Englewood Cliffs, NJ, 1995. 50. McDonough, R. and Whallen, A., Detection of Signals in Noise, 2nd ed., Academic Press, New York, 1995. 51. Tuzlukov, V., Signal Detection Theory, Springer-Verlag, New York, 2001. 52. Tuzlukov, V., Signal Processing Noise, CRC Press, Boca Raton, FL, 2002. 53. Crammer, H., Mathematical Methods of Statistics, Princeton University Press, Princeton, NJ, 1946. 54. Grenander, U., Stochastic Processes and Statistical Inference, Arkiv Mat, Uppsala, Sweden, 1950. 55. Rao, C., Advanced Statistical Methods in Biometric Research, John Wiley & Sons, New York, 1952. 56. Van Trees, H., Detection, Estimation, and Modulation Theory. Parts 1–4, John Wiley & Sons, New York, 2002–2003. 57. Maximov, M., Joint correlation of ﬂuctuating noise at the outputs of frequency ﬁlters, Radio Eng., No. 9, 1956, pp. 28–38. 58. Chernyak, Y., Joint correlation of noise voltage at the outputs of ampliﬁers with non-overlapping responses, Radio Phys. Electron., No. 4, 1960, pp. 551–561 (in Russian). 59. Tuzlukov, P. and Tuzlukov, V., Reliability increasing in signal processing in noise in communications, Automatized Systems in Signal Processing, Vol. 7, 1983, pp. 80–87 (in Russian). 60. Tuzlukov, V., Detection of deterministic signal in noise, Radio Eng., No. 9, 1986, pp. 57–60. 61. Tuzlukov, V., Signal detection in noise in communications, Radio Phys. Electron., Vol. 15, 1986, pp. 6–12. 62. Tuzlukov, V., Detection of deterministic signal in noise, Telecomm. Radio Eng., Vol. 41, No. 10, 1987, pp. 128–131. 63. Tuzlukov, V., Detection of signals with stochastic parameters by employment of generalized algorithm, in Proceedings of SPIE’s 1997 International Symposium on AeroSense: Aerospace/Defense Sensing, Simulation, and Controls, Orlando, FL, April 20–25, 1997, Vol. 3079, pp. 302–313. 64. Tuzlukov, V., Noise reduction by employment of generalized algorithm, in Proceedings of the 13th IEEE International Conference on Digital Signal Processing (DSP97), Santorini, Greece, July 2–4, 1997, pp. 617–620. 65. Tuzlukov, V., Detection of signals with random initial phase by employment of generalized algorithm, in Proceedings of SPIE’s 1997 International Symposium on Optical Science, Engineering, and Instrumentation, San Diego, CA, July 27–August 1, 1997, Vol. 3162, pp. 61–72. 66. Tuzlukov, V., Generalized detection algorithm for signals with stochastic parameters, in Proceedings of the 1997 IEEE International Geoscience and Remote Sensing Symposium (IGARSS’97), August 4–8, 1997, Singapore, pp. 139–141. 67. Tuzlukov, V., Tracking systems for stochastic signal processing by employment of generalized algorithm, in Proceedings of the 1st IEEE International Conference on Information Communications and Signal Processing (ICICS’97), Singapore, September 9–12, 1997, pp. 311–315. 68. Tuzlukov, V., A new approach to signal detection theory, Digital Signal Process. Rev. J., Vol. 8, No. 3, 1998, pp. 166–184.

Copyright 2005 by CRC Press

1598_book.fm Page 454 Monday, October 11, 2004 1:57 PM

454

Signal and Image Processing in Navigational Systems

69. Tuzlukov, V., Signal-to-noise ratio improvement under detection of stochastic signal using generalized detector, in Proceedings of the 1998 International Conference on Applications of Photonics Technology (ICAPT’98), Ottawa, Canada, July 27–30, 1988. 70. Tuzlukov, V., Signal processing in noise in communications: a new approach. Tutorial No. 3, in Proceedings of the 6th IEEE International Conference on Electronics, Circuits, and Systems, Paphos, Cyprus, September 5–8, 1999, pp. 5–128. 71. Tuzlukov, V., Detection of stochastic signals using generalized detector, in Proceedings of the 1999 IASTED International Conference on Signal and Image Processing, Nassau, Bahamas, October 18–21, 1999, pp. 95–99. 72. Tuzlukov, V., New remote sensing algorithms under detection of mineﬁelds in littoral waters, in Proceedings of the 3rd International Conference on Remote Sensing Technologies for Minefield Detection and Monitoring, Easton, Washington, D.C., May 17–20, 1999, pp. 182–241. 73. Tuzlukov, V., New remote sensing algorithms on the basis of generalized approach to signal processing in noise. Tutorial No. 2, in Proceedings of the 2nd International ICSC Symposium on Engineering of Intelligent Systems (EIS 2000), University of Paisley, Paisley, Scotland, June 27–30, 2000. 74. Tuzlukov, V., Detection of quasideterministic signals in additive Gaussian noise, News of the Belarusian Academy of Sciences, Ser. Phys.-Tech. Sci., No. 4, 1985, pp. 98–104 (in Russian). 75. Tuzlukov, V., Detection of signals with stochastic amplitude and random initial phase in additive Gaussian noise, Radio Eng., No. 9, 1988, pp. 59–61 (in Russian). 76. Tuzlukov, V., Interference compensation in signal detection for a signal of arbitrary amplitude and initial phase, Telecomm. Radio Eng., Vol. 44, No. 10, 1989, pp. 131–132. 77. Tuzlukov, V., The generalized algorithm of detection in statistical pattern recognition, Pattern Recognition and Image Analysis, Vol. 3, No. 4, 1993, pp. 474–485. 78. Tuzlukov, V., Signal-to-noise improvement in video signal processing, in Proceedings of SPIE’s 1993 International Symposium on High-Definition Video, Berlin, Germany, April 5–9, 1993, Vol. 1976, pp. 346–358. 79. Tuzlukov, V., Probability distribution density of background noise at the output of generalized detector, News of the Belarusian Academy of Sciences. Ser. Phys.Tech. Sci., No. 4, 1993, pp. 63–70 (in Russian). 80. Tuzlukov, V., Distribution law at the generalized detector output, in Proceedings PRIA’95, Minsk, Belarus, September 19–21, 1995, pp. 145–150. 81. Tuzlukov, V., Statistical characteristics of process at the generalized detector output, in Proceedings PRIA’95, Minsk, Belarus, September 19–21, 1995, pp. 151–156. 82. Tuzlukov, V., Signal ﬁdelity in radar processing by employment of generalized algorithm under detection of mines and mine-like targets, in Proceedings SPIE’s 1998 International Symposium on AeroSense: Aerospace/Defense Sensing, Simulations and Controls, Orlando, FL, April 13–17, 1998, Vol. 3392, pp. 1206–1217. 83. Tuzlukov, V., Employment of the generalized detector for noise signals in radar systems, in Proceedings of SPIE’s 2000 International Symposium on AeroSense: Aerospace/Defense Sensing, Simulations and Controls, Orlando, FL, April 24–28, 2000, Vol. 4048. 84. Tuzlukov, V., Adaptive generalized detector for unknown power spectral densities of noise, in Proceedings of the IEEE International Signal Processing Conference (ISPC’03), Dallas, TX, March 31–April 4, 2003.

Copyright 2005 by CRC Press

1598_book.fm Page 455 Monday, October 11, 2004 1:57 PM

Foundations of the Generalized Approach to Signal Processing in Noise

455

85. Tuzlukov, V., Signal detection in compound Gaussian noise: generalized detector, in Proceedings of the 3rd International Symposium on Image and Signal Processing and Analysis (ISPA’03), Rome, Italy, September 18–20, 2003. 86. Tuzlukov, V., Adaptive beam-former generalized detector, in Proceedings of IASTED International Conference on Signal Processing, Pattern Recognition, Applications (SPPRA 2003), Rhodes, Greece, June 30–July 2, 2003. 87. Kim, Y., Yoon, W-S., and Tuzlukov, V., Generalized approach to distributed signal processing with randomized data selection in wireless sensor networks, in Proceedings of the 1st International Conference on Information and Technology, Suwon, Korea, November 28–30, 2003, pp. 54–67. 88. Tuzlukov, V., Yoon, W.-S., and Kim, Y., Adaptive beam-former generalized detector in wireless sensor networks, in Proceedings of the IASTED International Conference on Parallel and Distributed Computing and Networks (PDCN 2004), Innsbruck, Austria, 2004, pp. 195–200.

Copyright 2005 by CRC Press

1598_C12.fm Page 457 Monday, October 11, 2004 3:22 PM

12 Theory of Space–Time Signal and Image Processing in Navigational Systems

12.1 Basic Concepts of Navigational System Functioning The main principle of navigational system functioning is the following. To deﬁne the location of the navigational object, for example, an aircraft, a moving image of the Earth’s surface or the totality of landmarks, relative to which the target moves, is formed on the board of the navigational object. An image of the same Earth’s surface or the totality of landmarks, which was obtained under the assumption that the target moves according to a predetermined trajectory, is formed in advance and is called the model image. The moving image is compared with the model image, and in doing so, a noncoincidence in the location of the totality of landmarks on the observation plane both for the moving image and for the model image is a measure of deviation of the aircraft ﬂight track from the predetermined or the required aircraft ﬂight track is characterized by mutual noncoincidence between the moving image and the model image relative to each other. Let us recall that a certain region of the given coordinate system is called the observation plane. In some cases, the region called the observation plane has a relative character. For example, in optical signal processing in navigational systems, the focus plane of the optical receiver is called the observation plane. Then, during the formation of the moving and model images as the reference totality, the domain of possible values of these readings related to the predetermined coordinate system is called the observation plane.1,2 The totality of landmarks used in solving navigational problems such as the Earth’s surface during navigation of aircraft ﬂight track, the image of the starry sky, and so on is observed, as a rule, in the background of interferences and noise generated by various sources. For this reason, the effectiveness of solving navigational problems is deﬁned by the quality of the moving image, i.e., by the quality of signal and image processing of information components of the moving image containing useful information regarding the totality of landmarks. It is obvious that the methods and techniques of space–time signal and image processing in navigational systems are deﬁned by the

Copyright 2005 by CRC Press

1598_C12.fm Page 458 Monday, October 11, 2004 3:22 PM

458

Signal and Image Processing in Navigational Systems

nature of the signals used. Because of this, it is worthwhile to carry out a brief investigation of the space–time ﬁelds and interference and noise generated by these ﬁelds before discussing the methods and techniques of signal and image processing in navigational systems. Generally speaking, any ﬁelds obtained by measurement of the characteristics of landmarks, the Earth’s surface, or the totality of other references, for example, the stars, can be considered as a source of information regarding the position of the navigational target. These characteristics, systematized in a ﬁxed way, can be presented as the ﬁeld of both the moving image and the model image. However, it is worthwhile to pay attention to the deﬁnition of those space–time signal processing methods and techniques that are widely used. Physical ﬁelds generating the space–time signals can be divided into natural and manmade ﬁelds. Anomalous magnetic ﬁelds, gravitation ﬁelds, the totality of readings of the Earth’s surface relief, optical ﬁelds, and heat and radio heat ﬁelds that are the consequence of the lighting of the Earth’s surface by the sun, moon, and stars, tectonic activity of the Earth, nonidentical radiant emittance of various nonhomogeneities of the Earth’s surface and ﬁelds, which are generated by human activity, are called the natural ﬁelds. In the case of human activity, the “natural ﬁeld” statement is conditional because this ﬁeld is not the result of purposeful human activity and can more rigorously be called the second-organized heat ﬁeld. The ﬁeld of radio signals caused by the radiation of radar, communications, and other signal processing systems can be considered the nonorganized ﬁeld. This ﬁeld is the secondary consequence of their functioning. Incidentally, the anomalous magnetic ﬁeld can be considered the secondary nonorganized ﬁeld because its character can be deﬁned by manmade metallic constructions, for example, a big bridge, dam, etc. The ﬁelds formed during the reﬂection of radiation energy directed to the navigational object are called manmade ﬁelds. They are formed by the radiation of the Earth’s surface by radar, laser, sonar, etc.3,4 The types of ﬁelds mentioned in the preceding text have various characteristics that can deﬁne the method of space–time signal and image processing. Moreover, the type of receiver or detector also inﬂuences the chosen method of the space–time signal and image processing. Let us explain this statement in more detail. As well known, the scanning of landmarks by, for example, the radar antenna of an aircraft navigational system can be both sequential and parallel. In the case of radar aircraft navigational systems, scanning and forming of an image of the two-dimensional (surface) target are carried out by the radar antenna, i.e., by the sequential displacement of the radar antenna. Thus, at each instant of time, the area of the scanned twodimensional (surface) target is deﬁned by the width of the directional diagram. This sweep can be partially realized due to the moving radar and antenna scanning in the direction that is orthogonal to the trajectory of the aircraft’s ﬂight (see Figure 12.1). An example of parallel space–time signal and image processing is the multielement optical receiver (matrix detector) in which an image of the scanned two-dimensional (surface) target is formed Copyright 2005 by CRC Press

1598_C12.fm Page 459 Monday, October 11, 2004 3:22 PM

Theory of Space–Time Signal and Image Processing in Navigational Systems

459

FIGURE 12.1 Orthogonal radar antenna scanning.

instantaneously. Evidently, in the ﬁrst case, the spatial image is transformed into the signal as a function of time and, in the second case, we have a spatial image of the two-dimensional (surface) target. It is clear that signal and image processing methods based on the generalized approach in the presence of noise5–8 are different for each case. Let us consider the sensor type stimulus of the considered ﬁeld on the signal and image processing algorithm. For instance, let us consider the radar navigational system. In the radar, the target return signal, as a rule, is a narrow-band stochastic process, though we would expect that it could be a high wideband signal process. The target return signal comes in at the input of the narrow-band linear tract or antenna of the navigational system. Resolution or accuracy of reproduction of the two-dimensional (surface) target is deﬁned by the wavelength of the searching signal and by the parameters and characteristics of the receiver or detector, ﬁrst and foremost by the width of the directional diagram. For the wavelength range in centimeter, the resolution of the aircraft radar navigational systems is a few hundred meters, and for the millimeter range, the resolution is a few tens of meters.9 In sonar navigational systems, the target return signal is characterized by mechanical oscillations. However, signal processing techniques in sonar navigational systems are very close to those in the radar because a sensor in the sonar ﬁeld transforms mechanical oscillations into electromagnetic oscillations similar to the radar video signal. In optical navigational systems, except in the case of laser and heat optic radiation of gases, the target return signal is a set of simple harmonic oscillations within the limits of the interval from 0.001 to 10 µm of electromagnetic waves and, therefore, signal and image processing under the use of resonance techniques and devices is a very complex problem in practice. The main type of receiver or detector in navigational systems operating in the optical range is the integral detector.10 Its basic principles are founded on the ability to transform optical signals into electromagnetic waves. Various techniques have been employed in the design of optical receivers and

Copyright 2005 by CRC Press

1598_C12.fm Page 460 Monday, October 11, 2004 3:22 PM

460

Signal and Image Processing in Navigational Systems

detectors. For example, a sensitive element can be placed in the focus of a concave mirror; this is the one-element receiver or detector. Sensitive elements can be presented in the form of a matrix, creating an image plane of the two-dimensional (surface) or three-dimensional (space) target; this is the multielement receiver or detector. Due to the low value of the electromagnetic wavelength, the resolution of optical navigational system is within the limits of the interval 1 to 10 m. The target return signal in laser navigational systems is different from that in optical navigational systems because it is a simple harmonic signal and can be subjected to coherent signal and image processing. For this reason, laser navigational systems have higher noise immunity.10 The correlation radius of the geophysical ﬁeld, which consists of magnetic and gravity ﬁelds, depends on the altitude, using which we can observe these ﬁelds. Even for low values of altitude, the correlation radius is within the limits of a few hundred meters.11 Moreover, there are no current methods and techniques to transform the geomagnetic ﬁeld into the electromagnetic wave in radar navigational systems, similar to the case in optical navigational systems where the target return signal is a function of the optical image in the coordinates of the observation plane. The existing sensors in geophysical ﬁelds allow measurement of a ﬁeld at the point corresponding to the location of the navigational object at the instant of measurement. A set of discrete readings obtained with the moving radar navigational system allows us to create a function, which corresponds to the cross section of the geomagnetic ﬁeld by the plane coinciding with the trajectory of the moving radar navigational system (see Figure 12.2). Comparing the observed moving image with the model images corresponding to possible variants of airborne ﬂight tracks and related to a given coordinate system, we can deﬁne the location of the navigational object based on analysis of the target return signals.12–15 It is obvious that in this case the signal and image processing of the moving image should be different from those of the space–time target return signals in the optical or radar ﬁelds. The moving image of the Earth’s surface relief is formed due to periodic changes in the aircraft’s ﬂight altitude. As an example, to solve the navigational problem of the moving ship, the sea depth is measured. In both the ﬁrst and second cases, the moving image is a function of time because of the moving radar navigational system. In this case, the signal and image processing is similar to the signal and image processing in time, as in the observation of the geophysical ﬁeld. Thus, methods and techniques of spatial target return signal processing can be classiﬁed into space signal and image processing, space–time signal and image processing, and signal and image processing in time. In space signal and image processing, each operation in deﬁning the elements composing the moving image is carried out at the same instants of time. In other words, elements of the moving image are processed in parallel. Therefore, space signal and image processing used in this case is a particular case of parallel processing of the target return signal containing all the Copyright 2005 by CRC Press

1598_C12.fm Page 461 Monday, October 11, 2004 3:22 PM

Theory of Space–Time Signal and Image Processing in Navigational Systems

461

FIGURE 12.2 Geomagnetic ﬁeld cross section.

necessary information. It is used when the sensor is the matrix receiver that is widely used in optics. In practice, these navigational systems are called raster or screen systems. The space–time signal and image processing is different in that one part of the moving image is subjected to parallel processing and the other part to sequential processing. An example is shown in Figure 12.3. An optical image of the observed area of the Earth’s surface is obtained by the mosaic linear sensor that is perpendicular to the axis of an aircraft. Thus, the moving image is obtained due to a shift of screen string formed at the receiver or detector output in navigational systems.

FIGURE 12.3 An example of space–time signal processing.

Copyright 2005 by CRC Press

1598_C12.fm Page 462 Monday, October 11, 2004 3:22 PM

462

Signal and Image Processing in Navigational Systems

The techniques of signal and image processing in time are characterized by sequential procedures of scanning of the observed moving image. They are realized by radar antenna scanning or the directional diagram of the sensor. Scanning of the observed area of the Earth’s surface can be done in various ways, for example, by spiral, row-frame sweep, moving radar, etc. However, in all cases, the observed moving image is a function of time. Signal and image processing of spatial moving images in time belongs in the technique of sequential information processing because each element of the moving image is deﬁned sequentially in time. Table 12.1 shows the different forms of the space–time signal and image processing applied to the main physical ﬁelds used in practice. It should be noted that the technique discussed takes an approximate character and shows the superiority of one or another kind of signal and image processing technique. The progress in signal and image processing of the moving image allows us to realize those algorithms and methods that have not been described previously. For instance, a mixed system of scanning the observed area of the Earth’s surface followed by the designing of the matrix image is widely used in practice. The element of the moving image in this navigational system is viewed on the exterior plane of the mirror by the infrared lens. The exterior plane of the mirror reﬂects the optical signal on the matrix element of the sensor receiver. The output signal of each element of the sensor receiver controls the photodiode of the indicator after ampliﬁcation. The exterior plane of the mirror is used for scanning a heat image, and the interior plane of the mirror is used to form a visible image in the plane of the photodiodes. In fast scanning by the mirror, this image can be subjected to parallel signal and image processing due to the inertness of photodiodes.16,17 TABLE 12.1 Forms of Space-Time Signal Processing Type of Field Geophysical Optical

Radar Radio heat

Copyright 2005 by CRC Press

Signal Processing Method Signal processing in time Signal processing in time Space–time signal processing Spatial signal processing Space–time signal processing Signal processing in time Space–time signal processing Signal processing in time

1598_C12.fm Page 463 Monday, October 11, 2004 3:22 PM

Theory of Space–Time Signal and Image Processing in Navigational Systems

463

12.2 Basics of the Generalized Approach to Signal and Image Processing in Time As noted earlier, the signal and image processing in time or the sequential signal and image processing of spatial signals is based on the transformation of the navigational ﬁeld into the signal as a function of time with further signal and image processing. The solution of navigational problems such as deﬁning the location of the navigational object is carried out under ambiguous conditions. The uncertainty is due to the following main factors: external interference, sensor noise, interference that can arise with the moving radar navigational system, indeﬁnite information of landmark coordinates, etc. These interferences caused by external noise and internal sensor noise are called signal interference or signal noise. Interferences acting on navigational systems are called direct interferences.18 Incorrect knowledge of the location of landmarks, i.e., the relative position of the navigational system in the coordinate system, is known as a priori uncertainty in the deﬁnition of the navigational object coordinates. For the reasons already noted, we can discuss only the estimation of the navigational object coordinates. We can consider the estimation to be optimal when it is the best in a deﬁnite sense. Thus, the main problem in navigational systems is deﬁning the optimal estimation of the navigational object coordinates, for example, deﬁning the true aircraft ﬂight track, using information including the signals generated by various physical sources. The deﬁnition of the optimal estimation of the navigational object coordinates is carried out in the following manner. Let us assume that there is information regarding the intensity and character of direct and signal interferences and we have information about the moving navigational object. Then, based on measurements that are available at this instant, we are able to truly estimate the navigational object coordinates. The process of deﬁning these estimations is signal and image processing. Let us consider the basic statements of classical signal processing theory within the limits of the Bayes approach to decision-making rules in the presence of uncertainty. The observed signal at the navigational system receiver input, in the general case, can be written in the following form: X = (Λ, n).

(12.1)

Here X is the m-dimensional vector of the observed signal at the receiver or detector input of the navigational system; Λ is the n-dimensional vector of parameters of coordinates of the navigational object; and n is the -dimensional noise vector ( ≤ m). We assume that the signal-to-noise ratio is known and already given. With the Bayesian approach, we consider that the joint probability distribution density f (Λ, n) is also already given. All information regarding the components of the vector of parameters Λ of the navigational object coordinates is

Copyright 2005 by CRC Press

1598_C12.fm Page 464 Monday, October 11, 2004 3:22 PM

464

Signal and Image Processing in Navigational Systems

included in the conditional probability distribution density f (Λ/X) that is the a posteriori probability distribution density of the vector Λ. In other words, f (Λ/X) contains all the information that may be used in the deﬁnition of the estimation of the navigational object coordinates. According to the Bayes theorem, we can write f ( Λ / X) = fps ( Λ ) =

fpr ( Λ ) ⋅ f (X / Λ ) f (X)

,

(12.2)

where fps(Λ) is the a posteriori probability distribution density; f (X/Λ) is the conditional probability distribution density deﬁning a statistic of the observed signal at the receiver or detector input of the navigational system; and fpr(Λ) is the a priori probability distribution density. The conditional probability distribution density f (X/Λ), considered as a function of Λ, is called the likelihood function (Λ). It contains all the information about the statistic of the observed signal samples at the input of the navigational system. These samples are processed to form the vector estimation Λ* of the vector of parameters of the navigational object coordinates. With ﬁxed values of the observed signal at the navigational system input, the likelihood function makes it possible for us to deﬁne the most probable value of Λ. In the deﬁnition of the vector estimation Λ*, its value tends to approach the optimal. The optimality of estimation is considered in accordance with the chosen criteria; let us recall the three widely used criteria.19 The ﬁrst criterion is that of the maximum a posteriori probability distribution density fps(Λ). With this criterion, the probability of the event that Λ = Λ* becomes maximal. The solution Λ* = fps(Λ) is called the most probable or unconditional estimation of the maximum probability distribution density fps(Λ). The second criterion is the minimum of the mean square error, i.e., min{<(Λ – Λ*)T O(Λ – Λ*)>} where <…> denotes the average in the statistical sense, and O is the positively deﬁned matrix introduced with the purpose of deﬁning the weight of the components of the error vector ∆Λ = Λ – Λ*. The maximum of the a posteriori probability distribution density coincides with the minimum of the mean square error in the case of some probability distribution densities, particularly the normal Gaussian probability distribution density. The solution Λ* = <Λ/X> is the estimation of the minimum variance. This estimation ensures the minimal width of the a posteriori probability distribution density. The third criterion is that in which the mean of the absolute error is minimized. The solution corresponding to the criterion can be written in the form Λ* = Me fps(Λ/X). Let us consider the particular case of the linear equation [see Equation (12.1)] as an example of the optimal estimation deﬁnition. For simplicity, we assume that the variables considered are not varied in time, i.e., we can write X = HΛ + n ; Copyright 2005 by CRC Press

(12.3)

1598_C12.fm Page 465 Monday, October 11, 2004 3:22 PM

Theory of Space–Time Signal and Image Processing in Navigational Systems

465

f (Λ, n) is the Gaussian probability distribution density that can be determined in the following form: f ( Λ , n ) = f ( Λ ) ⋅ f (n ) ;

(12.4)

A0 is the covariance matrix of Λ; = 0; = N, where N 1 ... 0 N = .......... .

(12.5)

0 ... N m Due to the stimulus of the additive noise, the probability distribution density f(X) is Gaussian with the mean <X> = HΛ0 and the covariance matrix equal to HA0HT + N. Let us deﬁne the a posteriori probability distribution density using the generalized approach to signal processing in noise.5–8 For this purpose, we will use the main statements of the generalized approach to signal processing in noise discussed in Chapter 11. In accordance with this, we have to form the additional reference signal η at the input of the navigational system. It is known a priori that the “no” information signal obtains in the additional reference signal η. The additional reference signal η, by its nature, is the noise or interference. Thus, we can write η = Q∗ (n ∗ ) ,

(12.6)

where = 0; = N*, and N 1∗ ... 0 N ∗ = .......... .

(12.7)

0 ... N m∗ The additional reference signal η is the m-dimensional vector, and n* is the -dimensional noise vector ( ≤ m). At the input of the navigational system signal X and additional reference signal η are observed and are uncorrelated processes by the initial premises of the generalized approach to signal processing in noise (see Chapter 11). For simplicity, we assume that the statistics N and N* are the same. Then, the a posteriori probability distribution density takes the following form: ln ( Λ∗ ) = ln [f ps ( Λ∗ )] =

Copyright 2005 by CRC Press

2X T N−1 ( Λ − Λ 0 ) − X T N−1 X + ηΤ N−1 η ( Λ − Λ 0 )T N−1 ( Λ − Λ 0 )

. (12.8)

1598_C12.fm Page 466 Monday, October 11, 2004 3:22 PM

466

Signal and Image Processing in Navigational Systems

Λ) is unknown, we can If the a priori probability distribution density fpr(Λ consider that it is uniform. In this case, the maximization of the a posteriori probability distribution density reduces to maximization of the likelihood function, i.e., to the deﬁnition of estimations that is optimal by the criterion of the likelihood function maximum. Let the signal X observed at the input of navigational system be the function of time deﬁned at the time instants reckoned over the same interval ∆t. Then the a posteriori probability distribution density can be written in the following form: fps ( Λ ) = f ( Λ / X) ,

(12.9)

where X = (X1, X2, …, Xp) is the sample of the signal vector at the input of the navigational system at the time instants ti(i = 1, 2, …, p). The deﬁnition of the optimal procedure of estimation of the vector Λ in the case of the continuous signal is carried out using the same rules as ∆t → 0. In doing so, a totality of samples of the signal observed at the input of the navigational system within the limits of the time interval [0, T] is replaced by the continuous process X(t). The likelihood function is replaced by the likelihood functional L( Λ ) = lim ( Λ ). ∆t → 0

(12.10)

Let us consider some examples. We assume that it is necessary to deﬁne the delay in the harmonic signal observed at the input of the navigational system. For simplicity, we assume that this delay can be considered as not varying in the time function during the time measurement interval and that the initial value of signal delay is equal to zero.

12.2.1 The Signal with Random Initial Phase Let us consider the working principles of the generalized detector shown in Figure 12.4. The initial conditions are the following. The signal at the input of the navigational system is observed within the limits of the time interval [0,T]. The signal is the narrow-band process with the given amplitude modulation law S(t) and phase modulation law Ψa(t). The random initial phase ϕ0 of the signal observed at the input of navigational system is distributed uniformly within the limits of the interval [–π, π] and is not time variant. The signal observed at the input of the navigational system has the following form: a(t) = S(t) cos[ω 0 t − Ψa (t) + ϕ 0 ] ,

Copyright 2005 by CRC Press

(12.11)

1598_C12.fm Page 467 Monday, October 11, 2004 3:22 PM

Theory of Space–Time Signal and Image Processing in Navigational Systems η(t)

AF

467

W (t )

X(t) PF Integrator

×

1

×

2

+

_

+

3 MSG

×

−

5

3 +

+ ×

+ +

2

4

1

+

× _

+

4

+

×

6

+

5

_

− Phase Control ω0τ = ϕ0

Amplitude Control AS(t ) = A*S *(t )

Phase Tracking System

Amplitude Tracking System

FIGURE 12.4 The generalized detector with tracking systems.

where ω0 is the carrier frequency of the signal a(t). The realizations of the input stochastic process correspond to variations of the random initial phase ϕ0 within the limits of the interval [–π, π]. Using the peculiarities of the generalized detector and the known amplitude and phase modulations of the signal at the input of the navigational system, we can construct a detector that is able to solve the problem of signal detection in the input stochastic process and that of deﬁnition and estimation of the random initial phase of the signal and, consequently simultaneously, signal delay. Because the model signal generated in the generalized detector is accurate with regard to the initial phase, it is not particularly a problem to design a device to measure and control the initial phase of the signal at the input of navigational system using the variation in the initial phase of the model signal at the output of the model signal generator (MSG) of the generalized detector. The process at the output of the preliminary ﬁlter (PF) of the generalized detector shown in Figure 12.4 takes the following form: X(t) = S(t) cos[ω 0 t + Ψa (t) − ϕ 0 ] + ξ 1 (t) cos[ω 0 t + υ(t)] .

Copyright 2005 by CRC Press

(12.12)

1598_C12.fm Page 468 Monday, October 11, 2004 3:22 PM

468

Signal and Image Processing in Navigational Systems

The process ξ(t) = ξ 1 (t) cos[ω 0 t + υ(t)]

(12.13)

is the narrow-band noise obeying, for simplicity of analysis, the Gaussian distribution law with zero mean and the ﬁnite variance σ n2 . The process at the output of MSG has the following form: a * (t) = S * (t) cos[ω 0 t + Ψa* (t)] .

(12.14)

Taking into account that the model a*(t) is not liable to match the signal a1(t) by the initial phase, we can write a * (t − τ) = S * (t − τ) cos[ω 0 (t − τ) + Ψa* (t − τ)] .

(12.15)

Because the amplitude S(t) and the phase Ψa(t) modulation laws of the signal at the input of the navigational system are known, we can always satisfy the conditions S * (t − τ) = S(t) and Ψa* (t − τ) = Ψa (t) .

(12.16)

Only under these conditions should the following analysis be carried out. Therefore, we drop the intermediate mathematics and consider only the main end results. The process at the output of the summator 1 takes the following form: Z1 (t) = 0.5S 2 (t) + S(t)ξ 1 (t) cos[Ψa (t) − υ(t) + ϕ 0 ] + 0.5[ξ 12 (t) − ξ 22 (t)] . (12.17) The processes at the outputs of the multipliers 1 and 2 have the following form: Z2 (t) = 0.5S 2 (t) cos[ω 0 τ − ϕ 0 ] + 0.5S(t)ξ 1 (t) cos[ω 0 τ − Ψa (t) + υ(t)]. (12.18) The process at the output of the summator 2 takes the following form: Z3 (t) = 0.5S2 (t){1 − cos[ω 0 τ − ϕ 0 ]} + 0.5S(t)ξ1 (t){2 cos[Ψ a (t) − υ(t) − ϕ 0 ] − cos[ω 0 τ − Ψ a (t) + υ(t)]} (12.19) + 0.5[ξ12 (t) − ξ 22 (t)].

Copyright 2005 by CRC Press

1598_C12.fm Page 469 Monday, October 11, 2004 3:22 PM

Theory of Space–Time Signal and Image Processing in Navigational Systems

469

The process at the output of the summator 3 takes the following form: Z4 (t) = S2 (t) cos[ω 0 τ − ϕ 0 ] – 0.5{S2 (t) − S(t)ξ1 (t){2 cos[ω 0 τ − Ψ a (t) + υ(t)] − 2 cos[Ψ a (t) − υ(t) − ϕ 0 ]}} + 0.5[ξ12 (t) − ξ 22 (t)] . (12.20) In the condition ω0τ = ϕ0 , the process at the output of the generalized detector has the following form: T

T

{∫ S (t) dt + ∫ [ξ (t) − ξ (t)] dt} .

Zgout (t) = 0.5

2

0

2 2

2 1

(12.21)

0

The condition ω0τ = ϕ0 is satisﬁed during the use of the phase tracking system (see Figure 12.4). The phase tracking system has the circuit that controls the initial phase of the model signal at the output of MSG of the generalized detector and fulﬁlls the required condition.

12.2.2 The Signal with Stochastic Amplitude and Random Initial Phase The signal with stochastic amplitude and random initial phase can be written in the following form: a(t , ϕ 0 , A) = A(t)S(t) cos[ω 0 τ + Ψa (t) − ϕ 0 ] ,

(12.22)

where ω0 is the carrier frequency of the signal a(t, ϕ0 , A); S(t) is the known modulation law of the amplitude of the signal a(t, ϕ0 , A); Ψa(t) is the known modulation law of phase of the signal a(t, ϕ0 , A); ϕ0 is the random initial phase of the signal a(t, ϕ0 , A), which is uniformly distributed within the limits of the interval [–π, π] and is not time variant within the limits of the time interval [0, T]; and A(t) is the amplitude factor, which is the random value and the function of time in the general case. The signal with stochastic amplitude and random initial phase at the output of PF of the generalized detector shown in Figure 12.4 for the case of the rapid ﬂuctuations of the amplitude factor A(t) within the limits of the time interval [0, T] takes the following form: a1 (t) = A(t)S(t) cos[ω 0 τ + Ψa (t) − ϕ 0 ] .

(12.23)

The model signal at the output of MSG of the generalized detector takes the form: Copyright 2005 by CRC Press

1598_C12.fm Page 470 Monday, October 11, 2004 3:22 PM

470

Signal and Image Processing in Navigational Systems a * (t) = A * (t − τ)S * (t − τ) cos[ω 0 (t − τ) + Ψa* (t − τ)] .

(12.24)

This written form is correct because the amplitude and phase modulation laws of the signal at the input of navigational system are known. Therefore, we can fulﬁll the conditions S * (t − τ) = S(t) and Ψa* (t − τ) = Ψa (t) .

(12.25)

With the arguments of Subsection 12.2.1, we can be assured that it is necessary to construct a detector that can solve the detection problem, deﬁne the time delay of the signals with stochastic parameters, and possess the amplitude and phase tracking systems. This detector is shown in Figure 12.4. In light of the main statements of Subsection 12.2.1, it is reasonably safe to suggest that the process at the output of the summator 1 takes the following form: Z1 (t) = 0.5 A 2 (t)S 2 (t) + A(t)S(t)ξ 1 (t) cos[Ψa (t) − υ(t) − ϕ 0 ] + 0.5[ξ 12 (t) − ξ 22 (t)] . (12.26) The process at the outputs of the multipliers 1 and 2 takes the following form: Z2 (t) = 0.5 A(t)A* (t)S2 (t) cos[ω 0 τ − ϕ 0 ] + 0.5 A* (t)S(t)ξ1 (t) cos[ω 0 τ − Ψ a (t) + υ(t)] .

(12.27)

The process at the output of the summator 2 takes the following form: Z3 (t) = 0.5S2 (t){ A2 (t) − A(t)A* (t) cos[ω 0 τ − ϕ 0 ]} + 0.5S(t)ξ1 (t){2 A(t) cos[Ψ a (t) − υ(t) − ϕ 0 ]

(12.28)

– A* (t) cos[ω 0 τ − Ψ a (t) + υ(t)]} + 0.5[ξ12 (t) − ξ 22 (t)] . The process at the output of the summator 3 takes the following form: Z4 (t) = A(t)A* (t)S2 (t) cos[ω 0 τ − ϕ 0 ] − 0.5{ A2 (t)S2 (t) – S(t)ξ1 (t){2 A* (t) cos[ω 0 τ − Ψ a (t) + υ(t)] – 2 A(t) cos[Ψ a (t) − υ(t) − ϕ 0 ]}} + 0.5[ξ12 (t) − ξ 22 (t)] .

Copyright 2005 by CRC Press

(12.29)

1598_C12.fm Page 471 Monday, October 11, 2004 3:22 PM

Theory of Space–Time Signal and Image Processing in Navigational Systems

471

Given ω0τ = ϕ0 , this process takes the following form: Z4 (t) = A(t)A* (t)S2 (t) − 0.5 A2 (t)S2 (t) + { A* (t) − A(t)}S(t)ξ1 (t) cos[Ψ a (t) − υ(t) − ϕ 0 ]

(12.30)

+ 0.5[ξ12 (t) − ξ 22 (t)] and under the additional condition A(t) = A*(t), the process at the output of the generalized detector has the form T

T

{∫ A (t)S (t) dt + ∫ [ξ (t) − ξ (t)] dt}.

Z (t) = 0.5 out g

2

2

0

2 2

2 1

(12.31)

0

Thus, with fulﬁllment of the conditions ω0τ = ϕ0

and

A(t) = A ∗ (t),

(12.32)

we are able to deﬁne the signal delay with high accuracy of estimation and minimum error; incidentally, the variance of error tends to approach zero under the condition T >> τc , where τc is the correlation length of the signal. The conditions in Equation (12.32) are fulﬁlled using the amplitude and phase tracking systems (see Figure 12.4). There are circuits that control both the amplitude and the initial phase of the model signal at the output of the model signal generator and provide fulﬁllment of the conditions in Equation (12.32). The results obtained indicate that we have a good chance of using the generalized detector in navigational systems in the signal processing of the target return signal with stochastic parameters, and that the estimation of the deﬁnition of the navigational object coordinates have a high accuracy. Let us proceed as follows: we construct amplitude and phase tracking systems that control the appropriate parameters of the model signal generated at the output of the model signal generator of the generalized detector. In principle, the construction of the amplitude and phase tracking systems is made possible by using the condition T >> τc . The amplitude and phase tracking systems may take various forms. One of the variants is shown in Figure 12.4. The detection performances of the generalized detector are shown in Figure 12.5 in comparison with the potential detection performances of the classical and modern signal processing theories.

Copyright 2005 by CRC Press

1598_C12.fm Page 472 Monday, October 11, 2004 3:22 PM

472

Signal and Image Processing in Navigational Systems PD 123

1.0

12 3

0.9

generalized detector with tracking system

0.8

generalized detector without tracking system Neyman-Pearson detector

0.7 0.6

1 2 3

0.5 0.4 0.3 0.2 0.1

q (dB) −10 0 2 4

6

8

10

12

13

14

15

16

17

18

FIGURE 12.5 Detection performances of the generalized detector: (1) Tβ = 1000; (2) Tβ = 100; (3) Tβ = 10; PF = 10–5.

12.3 Basics of the Generalized Approach to Space–Time Signal and Image Processing In practice, we deal with the spatial signals, not just time-dependent stochastic processes. The spatial signals are the radar or optical image of the Earth’s surface relief, functions deﬁning the magnetic or gravity ﬁelds of the Earth,12,15,18,20 and radio signals from various manmade sources, etc. The starry sky also forms a ﬁeld of spatial signals. The deﬁnition of real space–time stochastic processes is related to their presentation by the space–time signals.18,21 The application area of the space–time signals and images is very extensive, for instance, navigation, guidance, pattern recognition, pattern analysis, etc. Because space–time signal and image processing algorithms are similar to signal and image processing in time algorithms, and because space–time noise and interference also occur, the problem of optimal space–time signal and image processing arises in this case, too. Let us ﬁrst consider the widely used models of the space–time signals and images before considering the main results of the use of the generalized approach to space–time signal and image processing in noise. The analytical model of the space–time signals is deﬁned by the physical nature of these signals and problems, for solving which an observation is

Copyright 2005 by CRC Press

1598_C12.fm Page 473 Monday, October 11, 2004 3:22 PM

Theory of Space–Time Signal and Image Processing in Navigational Systems

473

s (x, y)

y

λy

λx

x

FIGURE 12.6 The parameter vector Λ deﬁnes the position of the signal.

carried out. One of the widely used models of the space–time signals corresponds to the surface potential of the space–time signal and image ﬁeld22,23 a( x , y , Λ , t) = a( x − λ x , y − λ y , t) and Λ = λ x , λ y

T

.

(12.33)

The position of the signal a(x, y, Λ, t) in the Cartesian coordinate system on the plane XOY is deﬁned by the parameter vector Λ (see Figure 12.6). The projections λx and λy of Λ deﬁne the shift of the signal along the corresponding axes. The ﬁeld model given by Equation (12.33) is called the ﬁrst-kind model. Its application area is too much varied. It is used in various navigational problems that are solved using optical and radar images of the Earth’s surface or individual landmarks; it is also used successfully in pattern recognition theory and image analysis. In a particular case, a totality of reference points, for example, a region of the starry sky, can be determined in the following form: M

a( x , y , Λ , t) =

∑ A δ(x − λ i

xi

, y − λ yi , t) ,

(12.34)

i=1

where δ(x) is the delta function. In some cases, it is necessary to know amplitude characteristics of the signal at a given point of 3-D space. This is characteristic of the air magnetic survey of the Earth’s anomalous magnetic gravity ﬁeld when it is necessary Copyright 2005 by CRC Press

1598_C12.fm Page 474 Monday, October 11, 2004 3:22 PM

474

Signal and Image Processing in Navigational Systems

to estimate the potential or other parameter of the ﬁeld.24–27 The ﬁeld potential a( x z ) for each point of the space x z is the linear integral operator of the surface potential a( x z ) =

∫ G(x , x ) ⋅ a(x ) dx ,

(12.35)

z

X

where G( x , x z ) is the known scalar function of vector arguments (the Green function) and x z is the function of time characterizing the moving position of the sensor and the law of scanning. If we can neglect the curvature of the Earth’s surface in the measurement of the geophysical ﬁeld, we can write

a( xz ) =

x3 z 2π

∞ ∞

∫∫

−∞ −∞

a( x1 , x2 ) dx1dx2 [( x1 − x1z )2 + ( x2 − x2 z )2 + x3 z ]2

(12.36)

.

where x z = ( x1z , x 2 z , x 3 z ) and x = ( x1 , x 2 )

(12.37)

are the Cartesian coordinate system shown in Figure 12.7. If we take into account the curvature of the Earth’s surface in spherical approximation, the spherical coordinates take the following form: x z = ( ϕ z , λ z , hz )

and

x = (ϕ , λ ),

(12.38)

where ϕz is the latitude, λz is the longitude, and hz is the altitude of sensor with respect to the Earth’s surface (see Figure 12.8). Then the potential can be written in the following form: a( x z ) =

(1 + h12 ) − 1 4π

∫∫ S

a(ϕ , λ ) cos ϕ dϕ dλ [1 + (1 + h12 ) − 2(1 + h1 ) cos Ψ]3

cos Ψ = sin ϕ sin ϕ z + cos ϕ cos ϕ z cos(λ − λ z ),

,

(12.39)

(12.40)

h1 = hR–1, where R is the radius of the Earth. In both cases, we believe that the ﬁeld characteristics are not varied in time to be true both for gravity and anomalous magnetic ﬁelds. This ﬁeld model, called the second-kind model, was designed by A. Krasovsky.12,28 To use the second-kind ﬁeld model it is necessary to deﬁne the Green function for each measurement. In particular, in the measurement of a single component of the anomalous magnetic ﬁeld strength or a single “gradient,” i.e., the

Copyright 2005 by CRC Press

1598_C12.fm Page 475 Monday, October 11, 2004 3:22 PM

Theory of Space–Time Signal and Image Processing in Navigational Systems x 3z

x 1z, x 2z, x 3z

x1 x 1z x 1z, x 2z

x2 x 1, x 2

x 2z FIGURE 12.7 The Cartesian coordinate system.

ϕz, λz, h

ϕ, λ

R

ϕz

λz FIGURE 12.8 The spherical coordinate system.

Copyright 2005 by CRC Press

λ

475

1598_C12.fm Page 476 Monday, October 11, 2004 3:22 PM

476

Signal and Image Processing in Navigational Systems

derivative of the second order

∂2 a ∂xzi ∂xzj

, there are 10 possible variants shown in

Table 12.2, where G(x, xz) is the Green function for the potential, and εTxi(i = 1, 2, 3) are the cosines between the vector of ﬁeld strength and the axes of the coordinate system. In this case, the Green function takes the following form: G( x , x z ) =

x3 z

.

2 π[( x1 − x1z ) + ( x 2 − 2 x 2 z )2 + x 32z ]3 2

(12.41)

In the case of the Earth’s sphere, we can write G( x , x z ) =

(1 + h1 )2 − 1 [1 + (1 + h1 )2 − 2(1 + h1 ) cos Ψ]3

.

(12.42)

TABLE 12.2 Possible Variants of Measurement of a Single “Gradient” Measured Value

Type of Operator

Resulting Kerner

Horizontal strength

∂ ∂x1z

∂ G( x, xz ) ∂x1z

Horizontal strength

∂ ∂x2z

∂ G( x, xz ) ∂x2z

Vertical strength

∂ ∂x3z

∂ G( x, xz ) ∂x3z

Total absolute value

3

∑

ε Txi

1

Copyright 2005 by CRC Press

∂ ∂xiz

x1x1

∂2 ∂x12z

x1x2

∂2 ∂x1z ∂x2z

3

∑ε

Txi

1

∂ G( x, xz ) ∂xiz

∂2 G( x, xz ) ∂x12z ∂

2

∂x1 z ∂x 2 z

G( x , x z )

x1x3

∂2 ∂x1z ∂x3z

∂2 G( x, xz ) ∂x1z ∂x3z

x2x2

∂2 ∂x22z

∂2 G( x, xz ) ∂x22z

x2x3

∂2 ∂x2z ∂x3z

∂2 G( x, xz ) ∂x2z ∂x3z

x3x3

∂2 ∂x32z

∂2 G( x, xz ) ∂x32z

1598_C12.fm Page 477 Monday, October 11, 2004 3:22 PM

Theory of Space–Time Signal and Image Processing in Navigational Systems

477

The space–time signal a(x, y, t) can be often expressed as the sum of the harmonic components. Each harmonic component can be considered as a simple harmonic ﬁeld, for which the inverse Fourier transform with respect to all frequencies is true:18,29 1 a( x , y , t) = ( 2 π )3

∞ ∞ ∞

∫ ∫ ∫ F(ω, ω , ω ) ⋅ e x

j ( ω x x + ω y y − ωt )

y

dω x dω y dω = −1 (ω , ω x , ω y ),

−∞ −∞ −∞

(12.43) where –1(ω, ωx, ωy) is the inverse Fourier transform; ∞ ∞ ∞

F(ω , ω x , ω y ) =

∫ ∫ ∫ A ( x , y) ⋅ e ω

− j ( ω x x + ω y y − ωt )

dx dy dt ,

(12.44)

−∞ −∞ −∞

where Aω(x, y) is the complex amplitude of the ﬁeld component at the given frequency ω; ωx is the spatial frequency along the axis x; and ωy is the spatial frequency along the axis y. Under these conditions, the space–time signal can be determined in the following form:

a( x , y) =

1 4π 2

∞ ∞

∫ ∫ F(ω , ω ) ⋅ e x

j(ω x x + ω y y )

y

dω x dω y = −1 {F(ω x , ω y )}, (12.45)

−∞ −∞

∞ ∞

F(ω x , ω y ) =

∫ ∫ a(x, y) ⋅ e

− j(ω x x + ω y y )

dx dy.

(12.46)

−∞ −∞

Let us recall that for a couple of the Fourier transforms to exist, the function a(x, y) must satisfy the Dirichlet conditions. The ﬁrst condition is: the observation interval can be divided on a ﬁnite number of intervals, within the limits of which the function a(x, y) is continuous and monotonic. The second condition is: the function a(x, y) should have limited numbers of ﬁnite breaks. Thus, there is one exponent with the weight coefﬁcient G(ωx, ωy) for each couple of values of the spatial frequencies ωx and ωy in the generalized sum corresponding to Equation (12.45). Let us consider this form of exponent for the given values of spatial frequencies. This exponent is the complex function similar to the Fourier transform of function in time. It is impossible to show this function graphically due to the double dependence on x and y. Therefore, the concept of zero-phase regions is introduced to guess the form of this function. Zerophase regions can be determined in the following form:

Copyright 2005 by CRC Press

1598_C12.fm Page 478 Monday, October 11, 2004 3:22 PM

478

Signal and Image Processing in Navigational Systems y

L

υ

θ

x

u

FIGURE 12.9 The nonzero phase region.

y = − (ux + 2 πn)υ −1 ;

u = 0.5π −1ω x ;

and

υ = 0.5π −1ω y . (12.47)

In accordance with Equation (12.47), the zero-phase regions can be represented by parallel lines with the distance between the lines equal to L = u 2 + υ 2 . The slope of these lines is deﬁned by the angle of orientation equal to ϑ = arctg uv (see Figure 12.9). When the spatial frequencies are high, the zero phase lines are located frequently. To understand this clearly, we should assume that the absolute value of the energy spectrum of spatial frequencies is equal to zero anywhere, excepting at the points (u1, υ1) and (–u2, υ1). In this case, the spatial signal can be approximated by the sum a( x , y) = 0.5 A ⋅ [e 2 jπ ( u1x + υ1y ) + e 2 jπ ( − u1x − υ1y ) ] ,

(12.48)

where A = const and is the real value that can be presented in the form of sine curve with the unit amplitude (see Figure 12.10). Peaks of this surface are the parallel lines similar to zero phase lines. Because, in the general case, the signal a(x, y, t) is a stochastic process, we should use the statistical characteristics, the mean, the variance, and the correlation function to deﬁne it. The correlation function for uniform — stationary by space — and stationary signals can be determined in the following form: R( ∆x , ∆y , ∆t) = < a( x + ∆x , y + ∆y , t) ⋅ a( x , y , t) > .

Copyright 2005 by CRC Press

(12.49)

1598_C12.fm Page 479 Monday, October 11, 2004 3:22 PM

Theory of Space–Time Signal and Image Processing in Navigational Systems

479

s

y

x FIGURE 12.10 Representation of the spatial signal.

In the case of the complex signals a(x, y, t), we can write R( ∆x , ∆y , ∆t) = < a( x + ∆x , y + ∆y , t + ∆t) ⋅ a˙˙( x , y , t) > .

(12.50)

Here, the symbol <.> denotes the mean in the statistical sense and the symbol a˙˙ denotes a complex conjugated value. The correlation functions given by Equation (12.49) and Equation (12.50) are called the space–time correlation functions. If we have a stationary space–time and uniform signal, which can be expressed as the product of the components of individual radiation sources, then we can write n

a( x , y , t) =

∑ a (x, y) ⋅ a (t), 1i

2i

(12.51)

i=1

where a1i(x, y) is the modulation law of the signal amplitude from the i-th source. If the modulation law of the signal amplitude from the i-th source is varied as a function of time in accordance with a2i(t), the space–time correlation function can be determined in the form of the product of the space and time components: R( ∆x , ∆y , ∆t) = R( ∆x , ∆y) ⋅ R( ∆t). Here Copyright 2005 by CRC Press

(12.52)

1598_C12.fm Page 480 Monday, October 11, 2004 3:22 PM

480

Signal and Image Processing in Navigational Systems

1 X →∞ XY Y →∞

R( ∆x , ∆y ) = lim

X Y

∫ ∫ a(x, y) ⋅ a(x + ∆x, y + ∆y) dx dy

(12.53)

− X −Y

is the space component and 1 T →∞ 2T

R( ∆t) = lim

T

∫ a(t) ⋅ a(t + ∆t) dt

(12.54)

−T

is the time component of the space–time correlation function. The formula in Equation (12.52) is true when stochastic variations of the space–time signal a(x, y, t) in time and on the plane XOY are independent. As in the case of signal processing in time, the theory of space–time signal processing possesses such statements as the power spectral density of spatial frequency S(ωx, ωy). In the case of the uniform ﬁeld, the power spectral density S(ωx, ωy) is deﬁned by the energy spectrum of the electromagnetic ﬁeld:30,31 S(ω x , ω y ) = lim| F(ω x , ω y )|2 . X →∞ Y →∞

(12.55)

The power spectral density of spatial frequencies of the uniform ﬁeld is related with the space–time correlation function by the Wiener–Heanchen theorem: ∞ ∞

S(ω x , ω y ) =

∫ ∫ R(∆x, ∆y) ⋅ e

− j ( ω x ∆x + ω y ∆y )

d( ∆x) d( ∆y),

(12.56)

−∞ −∞

R( ∆x , ∆y) =

1 4π 2

∞ ∞

∫ ∫ S(ω , ω ) ⋅ e x

y

j ( ω x ∆x + ω y ∆y )

dω x dω y = −1 [S(ω x , ω y )].

−∞ −∞

(12.57) For the correlation function given by Equation (12.52), the power spectral density S(ωx, ωy , ω) is deﬁned by the product of the angular power spectral density S(ωx, ωy) and the frequency power spectral density S(ω): S(ω x , ω y , ω ) = S(ω x , ω y ) ⋅ S(ω ) .

(12.58)

Optimization of the measurement of the navigational object parameter vector Λ for the space-time signals model of the ﬁrst kind observed in the presence of additive noise can be carried out based on the deﬁnition of the likelihood functional for the space–time signal and image processing.18,32 The Copyright 2005 by CRC Press

1598_C12.fm Page 481 Monday, October 11, 2004 3:22 PM

Theory of Space–Time Signal and Image Processing in Navigational Systems

481

η

t

tm t2 xk

t1 x

FIGURE 12.11 Space–time stochastic process η(x, y, t).

algorithm to solve this problem is as follows. First, we consider the multidimensional probability distribution density of the stochastic process ζ(x, y, t) that is discrete in space and time. The totality of the real values ζ(xk , yl , tm) where k, l, m = 1, 2, …, M, of the stochastic process ζ(x, y, t), which are called the samples, can be represented by readings of the speciﬁc realization obtained under the cross section of the stochastic process ζ(x, y, t) in the plane x = xk at the instant of the time tm (see Figure 12.11). For all values of xk , yl , and tm , the discrete readings ζ(xk , yl , tm) are the possible values of the stochastic process ζ(x, y, t) reckoned over the intervals ∆x, ∆y, and ∆t. The totalities of the probability distribution densities of these values for all M form the multidimensional probability distribution density f (M)[ζ(x, y, t)]. The limit of the multidimensional probability distribution density f (M)[ζ(x, y, t)] is deﬁned as the functional of the probability distribution density of the space–time stochastic process:18 f [ζ( x = x1 , y = y1 , t = t1 ); ζ( x = x1 , y = y1 , t = t2 ); … ; ζ( x = x1 , y = y1 , t = t M )];

(12.59)

f ( M ) [ζ( x , y , t)] = f [ζ( x = x1 , y = y1 , t = t1 )… ζ( x = x1 , y = y1 , t = t M ) …ζ( x = x 2 , y = y1 , t = t1 )… ζ( x = x 2 , y = y1 , t = t M )…ζ( x = x M , y = y M , t = t M )];

Copyright 2005 by CRC Press

(12.60)

1598_C12.fm Page 482 Monday, October 11, 2004 3:22 PM

482

Signal and Image Processing in Navigational Systems L(X) = lim f ( M ) [ζ( x , y , t)].

(12.61)

∆x → 0 ∆y → 0 ∆t → 0

Using the generalized approach to signal processing in noise,5–8 the likelihood functional in the case of the additive Gaussian noise can be written in the following form:

{

L(X) = C ⋅ exp − 0.5

∫ ∫ [2X S T

}

−1 ∗ 0

a − X T S−01 X + ηΤ S1−1 η] dx dy dt , (12.62)

X Y

where X = X( x , y , t) = X1 ( x , y , t)… X m ( x , y , t)

T

(12.63)

is the moving image or input stochastic process at the input of the navigational system that can contain the information signal; a ∗ = a ∗ ( x , y , t) = a1∗ ( x , y , t)… am∗ ( x , y , t)

T

(12.64)

is the model image; and η = η( x , y , t) = η1 ( x , y , t) … ηm ( x , y , t)

T

(12.65)

is the reference additional stochastic process at the input of the navigational system and it is the known a priori “no” information signal in the reference additional stochastic process; X is the observed region along the axis x; Y is the observed region along the axis y; S0 and S1 are the matrices of the power spectral densities in the sense of ordinary and spatial frequencies with the same statistics; and C is the normalized factor. Recall that the “white” Gaussian space–time noise is the Gaussian process with the non-time-variant power spectral density of components S0 or S1 in the sense of ordinary and spatial frequencies. The correlation function of the “white” Gaussian space–time noise can be written in the following form: R( ∆x , ∆y , ∆t) = S0 δ( ∆x , ∆y , ∆t). In the case of the one-dimensional (scalar) signal, we can write

Copyright 2005 by CRC Press

(12.66)

1598_C12.fm Page 483 Monday, October 11, 2004 3:22 PM

Theory of Space–Time Signal and Image Processing in Navigational Systems

483

{

L(X ) = C ⋅ exp − 0.5S0−1 ×

∫ ∫ ∫ [2X(x, y, t) ⋅ a (x, y, t) − X (x, y, t) + η (x, y, t)] dx dy dt}. 2

*

2

x y t

(12.67) Thus, the space–time signal and image processing using the generalized approach to signal processing in the presence of noise is to deﬁne the maximum of the integral I=

∫ ∫ [2X(x, y, t) ⋅ a (x, y, Λ , t) − X (x, y, t) + η (x, y, t)] dx dy, (12.68) ∗

∗

2

2

X Y

where X(x, y, t) is the moving image and a*(x, y, Λ*, t) is the model image. If the values X and Y are much more than the correlation length of the signal and the signal is the stationary stochastic process, the integral in Equation (12.68) matches the maximum of the likelihood functional. The deﬁnition of this maximum is carried out with the use of the generalized detector tracking systems discussed in Section 12.2. Thus, the discrimination characteristics are deﬁned along the corresponding axes: Ix =

∂ I ∂λ∗x

and

Iy =

∂ I. ∂λ∗y

(12.69)

Navigational systems realizing the generalized approach to signal processing in noise are called the correlation-extremal systems because Equation (12.68) allows us to deﬁne the maximum of the likelihood functional only under the main conditions of functioning of the generalized detector. The functions given by Equation (12.68) and Equation (12.69) are often called the criterial functions. The criterial functions given by Equation (12.69) are proportional to the measurement errors, i.e., Ix ≈

∂ 2 R( ∆x , ∆y , ∆t) ∂ 2 R( ∆x , ∆y , ∆t) ⋅ x and I ≈ ⋅ ∆y , ∆ y ∂( ∆x)2 ∂( ∆y )2 ∆x = 0 ∆y = 0 (12.70)

because the background noise in using the generalized approach to signal processing tends to approach zero in the statistical sense, and the noise component 2

∫ ∫ a (x, y, Λ , t)n(x, y, t)dxdy of the generalized detector corre*

∗

X Y

lation channel caused by the interaction between the model image and noise Copyright 2005 by CRC Press

1598_C12.fm Page 484 Monday, October 11, 2004 3:22 PM

484

Signal and Image Processing in Navigational Systems

and the random component 2

∫ ∫ a(x, y, Λ, t)n(x, y, t)dxdy of the generalized X Y

detector autocorrelation channel caused by the interaction between the moving image and noise are compensated by each other in the statistical sense (see Chapter 11). Correlation extremal receivers or detectors constructed according to the generalized approach to signal processing in the presence of noise may be used in navigational systems without tracking systems. The use of the generalized receivers or detectors of this form implies the employment of the signals with known amplitude-phase-frequency structure. Comparison of the incoming moving image at the input of the navigational system with the model image deﬁned by the form of the signal and a priori knowledge regarding the location of the signal on the observation plane allows us to estimate the true value of the parameter vector of Λ. In doing so, this value of the parameter vector Λ is automatically controlled. The main condition of the navigational system functioning without tracking devices is the correlation between the moving image X(x, y, t) and the model image a*(x, y, Λ*, t). In other words, we cannot recognize the information signal or image a(x, y, Λ, t) in the input moving image X(x, y, t). The use of the generalized detector with tracking systems (see Section 12.2) allows us to solve this problem and to recognize the type of the signal a(x, y, Λ, t) and to deﬁne and measure the parameter vector Λ. The criterial functions may be deﬁned in another manner. For example, in Equation (12.56) and Equation (12.57), which deﬁne the relationship between the space–time correlation function and the power spectral density of spatial frequencies, the criterial function takes the following form: 1 I= 4π 2

∞ ∞

∫ ∫ S(ω , ω , t) ⋅ e x

j ( ω x ∆x + ω y ∆y )

y

dω x dω y ,

(12.71)

−∞ −∞

where ∞ ∞

S(ω x , ω y , t) =

∫ ∫ R(∆x, ∆y, t) ⋅ e

− j ( ω x ∆x + ω y ∆y )

d( ∆x) d( ∆y).

(12.72)

−∞ −∞

For simplicity, we assume that the signal and noise are uncorrelated between each other in Equation (12.71) and the differentiation with respect to the parameters λ∗x and λ∗y is not taken into consideration. The discussion in the preceding text deﬁnes the so-called spectral technique of space–time signals processing. In accordance with this technique, the moving and model images are processed only in the spatial frequencies region. In this case, both the moving and model images are transformed in the power spectral densities of spatial frequencies according to Equation

Copyright 2005 by CRC Press

1598_C12.fm Page 485 Monday, October 11, 2004 3:22 PM

Theory of Space–Time Signal and Image Processing in Navigational Systems

485

(12.46). Therefore, the model signal a*(x – λx* , y – λy* , t) is deﬁned by F (ω x , ω y ) ⋅ e

− j ( ω x λ*x +ω y λ*y )

in accordance with the theorem of shift. The power

spectral density of the information signal in the moving image is deﬁned by F (ω x , ω y ) ⋅ e

− j ( ω x λ x +ω y λ y )

. Under these conditions, the criterial function given

by Equation (12.71) has the following form: ∞

I=

∞

∫ ∫ F(ω , ω ) ⋅ F (ω , ω ) ⋅ e ∗

x

y

x

− j ( ω x ∆x + ω y ∆y )

y

dω x dω y ,

(12.73)

− ∞ −…∞ ∞

∆x = λ x − λ∗x

and

∆y = λ y − λ∗y ,

(12.74)

where F*(ωx, ωy) denotes the complex conjugate Fourier transform. In other words, the signal processing given by Equation (12.71) and Equation (12.73) speciﬁes the transformation of the moving image into the power spectral density of spatial frequencies, the generation of the product of the complex conjugate power spectral densities of the moving image and model image, and fulﬁllment of the inverse Fourier transform to deﬁne the mutual correlation function between the information signal a(x, y, t) and the model signal Λ=Λ a*(x, y, Λ*, t) with further determination of its derivative at the point ∆Λ – Λ*. This technique is used successfully in optical navigational systems. We do not take into consideration the additive noise n(x, y, t) in Equation (12.71) and Equation (12.73). This was made with the purpose of stressing the external character of the criterial function. In practice, naturally, all spectral transformations are carried out with the moving image X(x, y, t). Because we are able to represent the moving image in the form of a set of harmonic components, the problem of the possible employment of various spatial frequency ﬁlters arises. It is well known5–8 that the generalized approach to signal processing in noise provides the deﬁnition of the mutual correlation function derivative of the moving and model images, which is equivalent to passing the moving image through the ﬁlter of the spatial frequencies with the pulse characteristic h( x , y ) =

∂a( x − λ∗x , y − λ∗y ) ∂λ∗x( y )

.

(12.75)

Actually, the criterial functions Ix and Iy are similar to the Duhamel integral, which is characteristic of the response of any system to external action. In this case, the derivative of the model image is considered as the pulse response of the preliminary ﬁlter of the generalized detector. We can use the preliminary ﬁlter of the generalized detector spatial frequencies with the pulse response h(x, y) = a*(x – λx* , y – λy* ), but in accordance with the Copyright 2005 by CRC Press

1598_C12.fm Page 486 Monday, October 11, 2004 3:22 PM

486

Signal and Image Processing in Navigational Systems

generalized approach to signal processing in the presence of noise, it is necessary to differentiate the signal at the output of the preliminary ﬁlter with respect to the variables λx* and λy* for each channel measuring the parameters λx and λy , respectively. For this purpose, we can use the preliminary ﬁlter of high spatial frequencies of the generalized detector. The stimulus of the preliminary ﬁlter of the generalized detector high spatial frequencies is analogous to spatial differentiation. Let us consider this problem in more detail using the example of the shift deﬁnition of the signal a(t) along the axis x, i.e., the example of the deﬁnition of the parameter λx. In accordance with Equation (12.69), the model image a*(x – λx* ) is the totality of the spatial waves of various lengths. The deﬁnite power spectral density corresponds to this representation. In the simplest case, when the stochastic process at the input of the navigational system has a single spatial harmonic in the condition ωx = ωx′ , we can write a* ( x − λ *x ) = 0.5 A ⋅ [e

j ( ω ′x x − ω ′y λ *x )

*

+ e − j(ω′x x − ω′xλ x ) ] .

(12.76)

In this case, the zero phase regions shown in Figure 12.9 have the form of an unlimited number of vertical parallel lines at the condition θ = 0° and are spaced by the distance x0 = (ωx′ )–1. The energy spectrum of spatial frequencies can be presented using the delta function Aδ(ωx – ωx′ ). In the general case, the space–time signal is formed by the sum of unlimited numbers of spatial harmonics, for example ∞

a (x − λ ) = *

* x

∑ 0.5A[e

∞

j ( ω xi x − ω xi λ *x )

+e

− j ( ω xi x − ω xi λ *x )

i=0

] = ∑ ci e jω

* xi λ x

,

(12.77)

i=−∞

where ci = 0.5A · ejωxix. Trigonometrically, we can write ∞

a* ( x − λ *x ) =

∑ A cos(ω x − ω λ ) . i

x

x

* x

(12.78)

i=0

The pulse response of the spatial frequency preliminary ﬁlter of the generalized detector has the following form: ∞

h( x) =

∑ cˆ ⋅ e i

− jω xi λ xi

and cˆi = 0.5 Ai ⋅ e − jω xi x .

(12.79)

i=−∞

Taking into consideration the averaging within the limits of the interval X >> x0i = ω2πxi , in terms of Equation (12.68) the criterial function has the following form: Copyright 2005 by CRC Press

1598_C12.fm Page 487 Monday, October 11, 2004 3:22 PM

Theory of Space–Time Signal and Image Processing in Navigational Systems ∞

IX = X

−1

∞

∫ ∑ ∑ 0.25A A ⋅ e i

k

− jω xik ∆x

487

dx

X i=−∞ k=−∞ ∞

+ 2 X −1

∫ ∑ 0.5A cos(ω i

xi

x − ω xi λ *x )ξ( x)dx

xi

x − ω xi λ *x )ξ( x)dx +

X i=−∞ ∞

– 2 X −1

∫ ∑ 0.5A cos(ω i

X i=−∞

∫ η (x) dx − ∫ ξ (x) dx, 2

2

X

X

(12.80) where the second term in Equation (12.80) is the noise component of the generalized detector correlation channel caused by the interaction between the model image and noise; the third term in Equation (12.80) is the random component of the generalized detector autocorrelation channel caused by the interaction between the information signal of the moving image and noise. The fourth and ﬁfth terms in Equation (12.80) are the background noise of the generalized detector. As discussed in Chapter 11, the second and third terms are compensated between each other in the statistical sense and the background noise [the fourth and ﬁfth terms in Equation (12.80)] tends to approach zero in the statistical sense. Based on the results discussed in Chapter 11, Equation (12.80) can be written in the following form: ∞

IX ≅

∑

0.25 Ai2 ⋅ e

− jω xi ∆x

∞

≅

i=0

∑ 0.5A cos(ω 2 i

xi

∆x),

(12.81)

i=0

where ∆x = λ – λx* and ωxik = ωxi – wxk. Thus, to obtain the criterial function in the form given by Equation (12.68), the preliminary ﬁlter of the generalized detector should be matched with the model image. The amplitude response of the amplitude-frequency characteristics of the generalized detector preliminary ﬁlter is the totality of the delta functions Aiδ(ωx – ωxi) (see Figure 12.12). Obviously, in practice we can only explain more close approximation of the amplitude-frequency response, for example, by the bell-shaped function (see Figure 12.13) ∞

∑ 0.5A ⋅ e i

−j

( ω x x − ω x λ 2x )2 i

i

2c

,

(12.82)

i=1

where c = const and the value of c is chosen as soon as it is low. Note that the use of spatial differentiation is equivalent to the procedure for emphasizing the contour lines that have been widely used in practice to process the optical images. Copyright 2005 by CRC Press

1598_C12.fm Page 488 Monday, October 11, 2004 3:22 PM

488

Signal and Image Processing in Navigational Systems

S (ω)

− ωx

1

− ωx

2

− ωx

3

0

ωx

1

ωx

2

ωx

3

ωx

FIGURE 12.12 Spectral characteristic of the high-frequency preliminary ﬁlter of the generalized detector.

S (ω)

− ωx

1

− ωx

2

− ωx

3

0

ωx

1

ωx

2

ωx

3

ωx

FIGURE 12.13 Spectral characteristic of the bell-shaped preliminary ﬁlter of the generalized detector.

In moving navigational systems, the zero-phase regions are shifted in parallel relative to the observed image of the Earth’s surface. In other words, the image of the Earth’s surface is formed due to the propagation of spatial harmonics with various spatial periods. The output voltage of the comb preliminary ﬁlter of the generalized detector is independent of the parallel shifts of the observed Earth’s surface image. For this reason, the preliminary ﬁlter of the generalized detector is kept matched with the model image. However, this preliminary ﬁlter is very critical with respect to the angle of

Copyright 2005 by CRC Press

1598_C12.fm Page 489 Monday, October 11, 2004 3:22 PM

Theory of Space–Time Signal and Image Processing in Navigational Systems

489

y

θ′

θPF θ x

FIGURE 12.14 Angles of mutual orientation.

mutual orientation θ′ = θ – θPF between the direction of the spatial wave normal θ and the axis θPF of the generalized detector preliminary ﬁlter (see Figure 12.14). If the pulse-transient function of the generalized detector preliminary ﬁlter can be considered as the model image or its derivative, the previous limitations tell us about the inadmissibility of the rotations between the information signal of the moving image and the model image. One way to cushion this requirement is to make the transfer function of the ﬁltering elements of the generalized detector preliminary ﬁlter more blurred, which ensures the invariance of the preliminary ﬁlter with respect to the angle of mutual orientation. In particular, in the one-dimensional case, i.e., θ = 0°, using n one-dimensional ﬁltering elements, the condition of invariance with respect to the allowable value of the mutual orientation angle can be deﬁned as32–35 n ≥ 0.5πθ′allow . According to the character of the additive noise n(x, y) at the input of the navigational system, we can use the generalized detector preliminary ﬁlter of various forms. For simplicity, we assume that the noise n1(x, y) is the Gaussian narrow-band process. Then, the criterial function for each estimation channel λx and λy is deﬁned by the derivative of the space–time correlation function of signal derivatives with the weight coefﬁcients that are inversely proportional to the spectral bandwidth of the noise n1(x, y) in the region of spatial and ordinary frequencies. In measuring the value of λx , this criterial function can be determined in the following form:

Copyright 2005 by CRC Press

1598_C12.fm Page 490 Monday, October 11, 2004 3:22 PM

490

Signal and Image Processing in Navigational Systems

I=

∂a ( x , y , Λ ) 1 ∫ ∫ {a(x, y, Λ) ⋅ ∂λ + 2α *

*

* x

X Y

2 x

×

∂ ∂a( x , y , Λ ) ∂a* ( x , y , Λ* ) ⋅ ∂x ∂x ∂λ *x

+

∂ ∂a( x , y , Λ ) ∂a∗ ( x , y , Λ∗ ) 1 ⋅ ⋅ ∂y ∂y 2α 2y ∂λ ∗y

[

]

[

(12.83)

]} dx dy,

−1 where α −1 x and α y are deﬁned by the components of the spatial correlation function of the additive noise n(x, y):

Rn ( ∆x , 0) = Rn (0, 0) ⋅ e − α x ∆x

and Rn (0, ∆y ) = Rn (0, 0) ⋅ e

− α y ∆y

. (12.84)

In an analogous way we can deﬁne the criterial function in measuring the value of λy . In estimating the one-dimensional parameter λ, Equation (12.84) can be written in the simplest form: I=

∂a ( x − λ ) 1 ∫ {a(x − λ ) ⋅ ∂λ + 2α *

x

X

* x

* x

2 x

⋅

* * ∂ ∂a( x − λ x ) ∂a ( x − λ x ) ⋅ ∂λ*x ∂x ∂x

[

]}dx. (12.85)

The physical form of Equation (12.83) and Equation (12.85) is, as noted, a differentiation that corresponds to the ﬁltering of high frequencies. Therefore, the spectral bandwidth of the additive noise is wider, the components of the high spatial frequencies of the moving image information signal are lower in value, and vice versa, i.e., if the main energy part of the additive noise n(x, y) is concentrated in the region of low frequencies, the most true information is contained in the high frequencies of the moving image information signal. If the space–time additive noise n(x, y, t)is a wide-band process so that α −x1 → 0 and α −y1 → 0, Equation (12.83) and Equation (12.85) have the form that is analogous to the “white” Gaussian noise considered earlier. In solving the navigational problems, the generalized approach to signal processing in the presence of noise allows us to use the following procedure. Let us assume that the intensity of the considered ﬁeld can be deﬁned as the totality of the values a(t, Λi). The additive noise is a stochastic process depending only on time, not on the observed space coordinates. The a priori information regarding the measured parameter vector Λi is determined by dΛ = Q(t , Λ ) + m(t), dt

Copyright 2005 by CRC Press

(12.86)

1598_C12.fm Page 491 Monday, October 11, 2004 3:22 PM

Theory of Space–Time Signal and Image Processing in Navigational Systems

491

where Q(t , Λ ) = q1 (t , Λ ), … , qn (t , Λ )

T

(12.87)

is the deterministic vector function of the parameter vector Λ and time; m(t) is the vector Gaussian noise with zero mean and given power spectral density matrix Mi ; and T

Λ = λ1 , λ 2 ,…, λ n .

(12.88)

In this case, the additive noise n(t) is a time stochastic process. Therefore, the likelihood functional in the case of the space–time Gaussian noise n(x, y, t) with zero mean and power spectral density C0 can be written in the following form:

{

L(X) = c ⋅ exp − 0.5

∫ ∫ ∫ [2X (x, y, t)C T

a ( x , y , Λ , t) − X T ( x , y , t)C −01 X( x , y , t)

−1 * 0

X Y T

}

+ ηΤ ( x , y , t)C1−1 η( x , y , t)]dxdydt . (12.89) In the case of the additive noise n(t) as a function of time only with zero mean and the power spectral density matrix N0, we can write

{

L(X) = c ⋅ exp − 0.5

∫ [2X (t)N T

−1 ∗ 0

}

a (t , Λ ) − X T (t)N−01 X(t) + ηΤ (t)N1−1 η(t)] dt .

T

(12.90) To realize in practice this space–time signal and image processing algorithm based on the generalized approach to signal processing in the presence of noise for solving navigational problems, we should have a special memory block to store both the map of the ﬁeld (the model image) and the map of the ﬁrst partial derivatives of the map of the ﬁeld (the ﬁrst partial derivatives of the model image).

Copyright 2005 by CRC Press

1598_C12.fm Page 492 Monday, October 11, 2004 3:22 PM

492

Signal and Image Processing in Navigational Systems

12.4 Space–Time Signal Processing and Pattern Recognition Based on the Generalized Approach to Signal Processing The pattern recognition process can be represented in the form of two steps. Measurements of deﬁnite estimations jointly with values of parameters belonging to the predetermined model image are carried out during the ﬁrst step. The extremal value of the criterion, which is chosen with the purpose of comparison, allows us to deﬁne with some probability the membership of the observed sample image in a deﬁnite group. Thus, two cases are possible. In the ﬁrst case, a priori knowledge about the observed image is very poor. Therefore, in the observation of the navigational object, we should compare the observed image (the moving image) with the great number of model images deﬁning the power of similarity. In the second case, we use the only model image, comparing it with the moving image using the totality of parameters. In this case, we are able to give the answer to the following question: Does the observed navigational object or the moving image belong to the same group as the model image or not? The pattern recognition problem is a more complex problem, i.e., in observing the moving image and using the only model image, we should answer the following question: Is there the expected image in the observed totality of the moving images or not? Thus, in the pattern recognition problem solving, we should carry out ﬁrst, the measurements, and second, the comparison between the moving and model images by analysis, and comparison of their parameters and features. Thus, the ﬁrst step precedes the second. In accordance with the generalized approach to signal processing in the presence of noise,5–8 the estimation of the moving image parameters is related to the use of the model image, which is a priori identical to the moving image. Actually, to measure the moving image parameters using the generalized approach to signal processing in the presence of noise, it is necessary to know the shape of the moving image (the observed object) and, consequently, we should deﬁne to which group the observed object (the moving image) belongs. To solve this problem, it is necessary to estimate the parameters and features of the moving image under the conditions of a priori uncertainty. Let us try to solve this illusory contradiction. The totality of estimations and distinctive features of the moving image (the recognized object) characterizes a certain point in the q-dimensional space. For example, the estimations of the two features x1 and x2 can be considered as coordinates on the plane XOY. The region of possible or allowable estimations of distinctive features forms the subspace of the moving image belonging to the deﬁnite group. The totality of other values of estimations forms the next region, and so on.

Copyright 2005 by CRC Press

1598_C12.fm Page 493 Monday, October 11, 2004 3:22 PM

Theory of Space–Time Signal and Image Processing in Navigational Systems

493

The widely used estimation of identity of images is the minimum of the difference between the vector X (the moving image) and the vector a* (the model image). If the vector components are dimensionless or have the same dimensions, we can use two nonweighted sum of difference squares between components of these vectors: ∆X = (X − a ∗ )T (X − a ∗ ).

(12.91)

The observed image (the moving image) belongs to the i-th group if the quadratic measure given by Equation (12.91) for the i-th model image is less than the same measure for any other model image, i.e., (X − a ∗i )T (X − a ∗i ) < (X − a ∗j )T (X − a ∗j )

for all

i ≠ j.

(12.92)

Each model image is on the q-dimensional sphere with the radius ρ, i.e., ρ2 = XTa*j is constant for all values of j. Therefore, the technique considered for the minimization of the distance between the vectors X and a*j is that the image characterized by the set of features x1, x2, …, xq belongs to the i-th group if X T a ∗i > X T a ∗j ,

i ≠ j.

(12.93)

The boundaries of the groups are fuzzy because we carry out measurements in the background of noise and interference, and allowable estimations for the given group of images are not exact due to a priori uncertainty. By this we mean that the end result of the pattern recognition problem has a probabilistic character as an estimation of distinctive features. Therefore, the statistical characteristics are used as the measure of convergence between the moving and model images. Under deﬁnite conditions, the correlation function between the vectors X and a*, i.e., R = <XTai* >, is the most convenient in this sense. Let us assume that there is a device measuring components of the vector X. Each component is a parameter of the observed object (the moving image). Let M be the possible image groups. The conditional probability distribution density fx|Hi(X|Hi) and a priori probability distribution density f(Hi) of the event Hi corresponding to the i-th image group are related to each group belonging to M. The pattern recognition problem is to deﬁne the technique of processing the observed vector X with the purpose of deﬁning the moving image group. As an example, let us consider the problem of detection of the signal with known shape AS(t) with unknown amplitude A in the background of the Gaussian noise. The input stochastic process has the form X(t) = AS(t) + n(t). The input stochastic process X(t) is limited by the spectral bandwidth ∆F of the generalized detector preliminary ﬁlter. The input stochastic process X(t) Copyright 2005 by CRC Press

1598_C12.fm Page 494 Monday, October 11, 2004 3:22 PM

494

Signal and Image Processing in Navigational Systems

is observed within the limits of the time interval [0, T]. We should make the decision a “yes” (hypothesis H1) or “no” (hypothesis H0) signal in the input stochastic process X(t). The conditional probability distribution density under the hypothesis H1 can be deﬁned by the likelihood functional

{

fX|H1 (X | H 1 ) = c ⋅ exp −

1 4σ n4

T

∫ [2X(t)a (t) − X(t)X(t) + η(t)η(t)] dt}, ∗

0

(12.94) where σ n2 is the variance of the noise at the input of the generalized detector preliminary ﬁlter, and c is the normalized cofactor. If we know the a priori probability distribution density fpr(A) of the amplitude of the signal, then the generalized detector determines the conditional probability distribution density f(X|H1) in the following form: fX|H1 (X | H 1 ) =

∫f

X|H1

(X | H 1 , A) ⋅ fps ( A) dA .

(12.95)

A

The statistic fX|H1(X|H1) at the output of the generalized detector is compared with the threshold. If the statistic exceeds the threshold, we make the decision a “yes” information signal in the moving image and vice versa. Thus, the pattern recognition problem solving requires the preliminary estimation of the amplitude of the signal that should be taken into consideration in determining the probability distribution density fX|H1(X|H1) given by Equation (12.95). If the preliminary ﬁlter of the generalized detector has the spectral bandwidth ∆F, then the input stochastic process can be represented as the vector consisting of 2T∆F readings:5–8 X( 2 ∆1F ) X=

X( ∆1F )

.

(12.96)

X(T ) The conditional probability distribution density under the hypothesis H1 takes the following form: fX|H1 (X | H 1 , A) =

Copyright 2005 by CRC Press

2X T C 0−1 AS − X T C 0−1 X + ηΤ C 1−1 η 1 ⋅ − exp , 8σ n4 ∆F (8σ n4 ∆F )T∆F (12.97)

{

}

1598_C12.fm Page 495 Monday, October 11, 2004 3:22 PM

Theory of Space–Time Signal and Image Processing in Navigational Systems

495

where S( 2 ∆1F ) S=

S( ∆1F )

.

(12.98)

S(T ) For simplicity, we assume that the additional ﬁlter has the same spectral bandwidth ∆F in value. In observing the spatial signals a(x, y) and space–time signals a(x, y, t), the pattern recognition principles are the following.36,37 We should ﬁnd the region on the observed plane XOY in which the function of intensity is similar to the predetermined function of intensity called the model function. In this case, the criterion of similarity is the quadratic measure given by Equation (12.91), which can be reduced to the correlation function between the moving and model images R = < a( x , y) ⋅ a∗ ( x − λ x , y − λ y ) >

(12.99)

in the case of the Gaussian probability distribution density of the noise where the parameters λx and λy deﬁne the shift between the model image relative to the moving image belonging to the same group of images. The maximum of the correlation function given by Equation (12.99) corresponds to the complete matching between the model image and the moving image. A position of the model image in the coordinate system XOY is known a priori (see Figure 12.15). Thus, the deﬁnition of the maximum of the correlation function given by Equation (12.99) allows us to solve both the pattern recognition problem and to deﬁne and measure the position of the moving image in the coordinate system XOY. Let us demonstrate this. Let the information space–time signal ai(x – λ, t) belonging to the i-th group be observed in the background of the space–time noise n(x, t) with zero mean and power spectral density C0. For simplicity, we do not take into consideration the coordinate y of the coordinate system XOY. We need to deﬁne what is the group p, to which the moving image belongs. In the general case, the pattern recognition procedure can be carried out using the p-channel generalized receiver or detector. The conditional probability distribution density of the statistic at the generalized detector output under the hypothesis Hi corresponding to the presence of the i-th signal (i = 1, 2, …, p) can be determined in the following form:5–8 fX|Hi (X | H i ) =

Copyright 2005 by CRC Press

∫f

X|H i

[(X | H i )|λ i ] ⋅ f (λ i ) dλ i

(12.100)

1598_C12.fm Page 496 Monday, October 11, 2004 3:22 PM

496

Signal and Image Processing in Navigational Systems y s (x, y, Λ) s (x, y, Λ*)

λy λ*y

x λx λ*x FIGURE 12.15 The displacement of the model image in the coordinate system XOY.

for each channel of the generalized detector, where X = X( x , t) = ai ( x − λ i , t) + n( x , t),

(12.101)

< n( x + ∆x , t + ∆t) ⋅ n( x , t) > = C0 δ( ∆x , ∆t),

(12.102)

f(λi) is the a priori probability distribution density of the parameter λi . Equation (12.102) is true only at the inputs of the preliminary and additional ﬁlters because, after passing the preliminary and additional ﬁlters of the generalized detector, the Gaussian noise will be limited by the spectral bandwidth. The statistics at the output of each i-th channel are compared with the purpose of deﬁning the greatest value and to make a decision regarding the hypothesis H1 for the i-th channel. Thus, the pattern recognition problem is solved by the deﬁnition of the conditional probability distribution density fX|Hi[(X|Hi)|λi] at the output of the generalized detector, which can also be called the likelihood functional. In the case of the space–time additive noise, we can write fX|Hi [(X |H i )|λ i ] = c

{

× exp −

1 8σ n4

∫ ∫ [2X(x, t)a (x − λ , t) − X(x, t)X(x, t) + η(x, t)η(x, t)] dx dt}. ∗

∗ i

X T

(12.103)

Copyright 2005 by CRC Press

1598_C12.fm Page 497 Monday, October 11, 2004 3:22 PM

Theory of Space–Time Signal and Image Processing in Navigational Systems

497

If the dimensions of the observed region X are much more than a correlation length of the signal ai(x, t), we can write fX|Hi [(X |H i )|λ i ] = c ⋅ exp{−

[2 Rii∗ − Rii + 2 Ri∗n − 2 Rin + Rn ]} ,

1 8 σ n4

(12.104) where Rii∗ ≅

1 XT

∫ ∫ a (x − λ , t) ⋅ a (x − λ , t) dx dt;

(12.105)

∫ ∫ a (x − λ , t) dx dt;

(12.106)

i

2 Rin ≅

Rn ≅

∗ i

X T

Rii ≅

2 Ri∗n ≅

∗ i

i

1 XT

2 i

i

X T

1 XT

∫ ∫ 2a (x − λ , t) ⋅ ξ (x, t) dx dt;

(12.107)

1 XT

∫ ∫ 2a (x − λ , t) ⋅ ξ (x, t) dx dt;

(12.108)

1 XT

∫ ∫ [η (x, t) − ξ (x, t)] dx dt;

(12.109)

∗ i

∗ i

i

X T

i

i

i

X T

2 i

2 i

X T

Rii* is the correlation function between the information signal ai(x – λi, t) in the moving image Xi(x, t) and the model image ai* ( x − λ*i , t) at the output of the i-th channel of the generalized detector; Rii is the correlation function of the information signal ai(x – λi, t) in the moving image Xi(x, t); Ri*n is the correlation function of the correlation channel noise component caused by the interaction between the model image ai* ( x − λ*i , t) and the noise ξi(x, t) at the output of the i-th channel; Rin is the random component correlation function of the autocorrelation channel caused by the interaction between the information signal ai(x – λi, t) in the moving image Xi(x, t) and the noise ξi(x, t) at the output of the i-th channel; Rn is the correlation function of the background noise; ξi(x, t) is the noise at the output of the preliminary ﬁlter; and ηi(x, t) is the noise at the output of the additional ﬁlter; the noise ξi(x, t) and ηi(x, t) are uncorrelated (see Chapter 11). Because the signal and noise are uncorrelated at the input of the generalized detector and the difference between the correlation functions Ri*n and Rin tends to approach zero in the statistical sense if the coefﬁcient of correlation between the model image ai* ( x − λ*i , t) and the information signal ai(x – λi, t) in the moving image Xi(x, t) is equal to unity, and the background

Copyright 2005 by CRC Press

1598_C12.fm Page 498 Monday, October 11, 2004 3:22 PM

498

Signal and Image Processing in Navigational Systems

noise at the output of the generalized detector tends to approach zero in the statistical sense, the probability distribution density fX|Hi[(X|Hi)|λi] is deﬁned by the accuracy of the position deﬁnition λi of the information signal in the moving image Xi(x, t). Actually, under these conditions, we can write fX|Hi [(X |H i )|λ i ] = c ⋅ exp{−

1 8 σ n4

[Rii∗ − Rii ]} .

(12.110)

The exponent in Equation (12.110) varies monotonically depending on values of Rii* and Rii. Therefore, the correlation functions Rii* and Rii deﬁne completely the character and behavior of the probability distribution density fX|Hi[(X|Hi)|λi]. In the condition i = i*, i.e., when the moving image and the model image belong to the same group and an error ∆λ = λi – λi* of the information signal position measurement in the moving image Xi(x, t) is close to zero, the probability of true recognition is maximal. Actually, due to low errors during the model image tuning and at the condition i = i*, we can assume that Rii∗ − Rii ≅ Rii′ ∗ ∆λ,

(12.111)

where Rii*′ is the derivative of the correlation function at zero; ∆λ = λi – λi* ; and λi* is the estimation of the moving image shift. Thus, fX|Hi[(X|Hi)|λi] is the a posteriori probability distribution density of shift estimation of the observed i-th moving image Xi(x, t). The generalized receiver has p channels. Each channel uses its own i-th model image with the initial tuning deﬁned by the expected position of the i-th moving image on the observation plane. If the i-th information signal is in the moving image, then the i-th channel of the generalized receiver deﬁnes the optimal by the maximum a posteriori probability distribution density estimation of its position λi* . The multichannel generalized receiver is analyzed in more detail in Tuzlukov.8 Thus, the composition of the p-channel generalized receiver allows us to solve the pattern recognition problem in the following manner. The moving image X( x , t) comes in at each input of the p-channel generalized receiver. The moving image X( x , t) is the discrete stochastic process. There is a discrete model image for each channel of the p-channel generalized receiver. The position of the model image at the initial instant of time is deﬁned by the a priori knowledge and the estimation λi* obtained earlier. If for the i-th channel of the generalized detector we obtain the less value of σ ii2 ∗ , where σ ii2∗ = <(λ i − λ ∗i )2 >, we can determine the “yes” information signal in the moving image, and the information signal belongs to the same group as the model image. In the sequential pattern recognition technique, we can use the one-channel generalized detector with a set of model images. In this case, we should have p model images. This technique is used in a small sample of moving images. If the sample is large, the computing cost is very high. Copyright 2005 by CRC Press

1598_C12.fm Page 499 Monday, October 11, 2004 3:22 PM

Theory of Space–Time Signal and Image Processing in Navigational Systems

499

12.5 Peculiarities of Optical Signal Formation Let us consider the peculiarities of optical signal formation in navigational systems in the deﬁnite order caused by speciﬁc conditions. The input stochastic process is formed in the space called the object space in the form of characteristics of the landmark totality. The Earth and sea surfaces, the cosmic space, etc., can be considered the object space. The information signals generated by the object space come in at the input of the navigational system. The information signals are transferred by the channel that can be the propagation medium of the optical signals. The input information signal is processed by the optical receiver containing the preliminary ﬁlter for the ordinary frequency range and the spatial frequencies. The optical receiver forms the object images within the limits of vision ﬁeld, i.e., the moving image. The moving image is reproduced by the receiver, the form of which deﬁnes the procedure of signal processing with the use of the generalized approach in the presence of noise. Using this sequence of the procedure, let us consider the process of optical signal formation. Depending on the spectrum range in which the transmitter or source generates signals, we use the energy and photoelectric characteristics to deﬁne the peculiarities of the optical signal source. The transition from energy characteristics to photoelectrical characteristics is not difﬁcult.38 The procedure of optical bearing is based on the use of image contrast of the observed objects (landmarks) in the background of noise. The image contrast exists because they have features that radiate or rereﬂect the energy of electromagnetic waves into the optical wave range to a greater or lesser degree in comparison with sources of interference and noise that are around these objects. The summation of the radiation of the object is caused by two components: internal heat radiation and reﬂected radiation from natural or manmade sources (sun, moon, stars, the Earth, the atmosphere, manmade lighting, etc.). The type of these sources deﬁnes the character of the natural or manmade optical ﬁeld. However, it should be noted that in the majority of cases, the summing ﬁeld is formed simultaneously under the stimulus of natural and manmade sources. This applies restrictions on signal processing in a wide range of optical wavelengths. Therefore, we should use additional techniques such as spectral selection. The range 0.2 … 3 µm is traditionally recommended for employment in reﬂected and scattered solar radiation. The range 10 … 12 µm is recommended for employment in radiation of the Earth’s surface. Evidently, a redistribution of energy between own and reﬂected radiation can be carried out within the limits of short time intervals and sufﬁciently wide limits.39,40 All these components of the summing object radiation depend essentially on the shape, construction, material, orientation, position on terrain, regime of functioning, etc., of the object. In addition, a state of the environment, location of external sources, and other factors play a large role. These factors

Copyright 2005 by CRC Press

1598_C12.fm Page 500 Monday, October 11, 2004 3:22 PM

500

Signal and Image Processing in Navigational Systems

generate a high deviation of radiation characteristics that can be deﬁned using the theory of statistical decision making. Thus, the problem of deﬁnition of manmade objects or landmarks can be simpliﬁed due to their deﬁnite uniformity of shape, dimensions, and material with which these objects are produced. Unlike natural noise and interferences, which can be considered as a background, manmade object-landmarks possess a low deviation of various radiation components. The dependence on direction and power of the wind and seasonal conditions is less for manmade object-landmarks. Incidentally, the direction and power of the wind and seasonal conditions can signiﬁcantly change the background, for example, orientation and color of tree leaves, the slope of plant stems, sea and lake surfaces, etc. The internal radiation of objects is a function of the surface temperature and physical features. This radiation can be both coherent and noncoherent. Coherent radiation is a peculiarity of electromagnetic radiation that keeps the difference in phases within the limits of the time interval that is required to detect and measure this difference. Evidently, in the majority of cases, any navigational system processes noncoherent signals. The most informative characteristic of internal noncoherent radiation is the spectral energy brightness of the heated object. This characteristic can be determined in the following form: λ = ε λ ⋅

1.19 ⋅ 10 4 λ5 (e

14388 λT

− 1)

,

(12.112)

where T is the temperature of the radiation source, λ is the wavelength, and ελ is the spectral power of the blackness of the radiating surface at the given temperature and deﬁnite bearing. To deﬁne the coherent radiation of the object, it is necessary to take into consideration its interference and diffracting characteristics. Characterizing the efﬁciency of internal radiation of the heat sources, we can deﬁne three kinds of emitters: blackbody, gray emitter, and selective emitter. The parameter ελ deﬁnes the efﬁciency of radiation for the given wavelength. Sometimes, the parameter ελ is called the coefﬁcient of radiation. The absolute blackbody has a black power ελ = 1 for the whole range of the wavelength. In the case of the gray emitter, we obtain ελ = const <1 within the limits of the deﬁnite range of the wavelength. In the case of the selective emitter, we have 0 ≤ ελ<1, and the parameter ελ is an unambiguous function of λ of any kind. Speaking rigorously, the degree of blackness of real objects ελT is always a function of the wavelength and temperature. For this reason, the constant value of ελ can be deﬁned within the deﬁnite limits of change of λ and T. This function depends also on the angle of the vision ﬁeld. However, these angles are very low in value in navigational systems of any kind, as a rule. Therefore, we can consider the parameter ελ as independent of the angle of vision ﬁeld.

Copyright 2005 by CRC Press

1598_C12.fm Page 501 Monday, October 11, 2004 3:22 PM

Theory of Space–Time Signal and Image Processing in Navigational Systems

501

The working principles of the majority of objects are based on the use of energy equipment. A large amount of heat energy is released as a result of the functioning of energy equipment. A part of this heat energy is discharged into the environment. For this reason, all objects have one more peculiarity: the internal radiation of the vapor phase that is often called a ﬂare (the vapor stream of reactive engines, ship pipes, explosion stacks, etc.) is added to the internal surface radiation of the object. The dependence of the spectral degree of blackness on the wavelength of those vapor sources takes the form of oscillations unlike with the smooth character of solid metallic objects. The reﬂection features of objects depend on the relationships between the dimensions of structure heterogeneities of the object surface and the wavelength of the incident radiation. If these dimensions are much less than the wavelength, we can consider that there is a mirror reﬂection. If these dimensions are commensurable with the wavelength, there is a scattered reﬂection. The main characteristic of reﬂection features is the reﬂection coefﬁcient µ0 characterized by the ratio between the radiation stream reﬂected by the object surface and the incident radiation stream on the object surface. In mirror reﬂection from the ﬂat surfaces of objects, the coefﬁcient of reﬂection can be deﬁned as the ratio between the surface brightness after reﬂection and the initial surface brightness under the condition that a space angle, within the limits of which the incident radiation is propagated, is kept constant after radiation. This condition is not satisﬁed with scattered reﬂection when the space angle, within the limits of which the reﬂected radiation is propagated, is greater than that within the limits of which the incident radiation is propagated.41 The limit of the space reﬂection angle for a ﬂat surface is equal to 2π. In this case, the surface is called a diffuse reﬂection surface and the reﬂection is called the diffuse or Lambert reﬂection. The brightness of this surface is the same in value for all directions of radiation propagation, i.e., it is independent of both the angle θ in the meridian plane and the azimuth angle ϕ of the viewﬁnder plane. In addition, the brightness is independent of both the antiaircraft angle θ′ and the azimuth angle ϕ′ deﬁning the position of the radiation source. The brightness of the ideal scattered surface is functionally related to its illumination by an extraneous radiation source E(θ′, ϕ′) and can be determined in the following form: 0 (θ′, ϕ′) =

E(θ ′ , ϕ ′) . π

(12.113)

Thus, a scattered radiation by its character can be uniform and nonuniform if the brightness distribution in space depends on the direction of the viewﬁnder and location of the radiation source. The coefﬁcient of the brightness characterizes this feature of scatterers. The coefﬁcient of brightness µb(θ, ϕ, θ′, ϕ′) is the ratio between the brightness characteristic of the object surface region along the given direction and

Copyright 2005 by CRC Press

1598_C12.fm Page 502 Monday, October 11, 2004 3:22 PM

502

Signal and Image Processing in Navigational Systems

the brightness 0 (θ′, ϕ′) of the ideal scattering diffuse surface having the coefﬁcient of reﬂection equal to unity, i.e., the absolutely “white” surface, which is in the same conditions of brightness and observation, µ b (θ , ϕ , θ ′ , ϕ ′ ) =

(θ, ϕ , θ ′ , ϕ ′) . 0 (θ ′ , ϕ ′ )

(12.114)

Therefore, the brightness of the real object surface along the direction of observation or the viewﬁnder is given by (θ, ϕ, θ′, ϕ′) =

E(θ ′ , ϕ ′) ⋅ µ b (θ , ϕ , θ ′ , ϕ ′ ) . π

(12.115)

The speciﬁc conditions of functioning of the navigational system and the feature of objects in absorbing and reﬂecting the radiation selectively allow us to deﬁne the spectral reﬂection ability by a set of coefﬁcients of the spectral brightness µbλ(θ, ϕ), which are ratios of the spectral brightness of the objects and the spectral brightness of the ideal scattering surface under the same conditions of illumination and observation, i.e., µ bλ (θ, ϕ) =

λ (θ , ϕ , θ ′ , ϕ ′ ) . 0 λ (θ ′ , ϕ ′ )

(12.116)

Consequently, the coefﬁcients of brightness are the function of the zenith and azimuth angles in the deﬁnition of the radiation source position, as a rule of the sun. This function cannot be deﬁned in analytical form and is presented, as a rule, in the diagrammatic or graph form using polar diagrams or graphs, the radius-vector lengths of which are proportional to the values of the coefﬁcients of brightness in corresponding directions. These diagrams or graphs are usually called the indicatrices of reﬂection or scattering. As a rule, the indicatrices of brightness are plotted in the form of cross sections of the indicatrices of reﬂection using polar or Cartesian coordinate systems. The brightness and, consequently, the coefﬁcient of brightness are the functions of the coordinates x, y, z of the radiation point in the object space, the time t, and the wavelength λ. This fact causes some difﬁculties in describing radiation models. However, the conditions of optical-electronic equipment employment in various navigational systems carrying out speciﬁc functions allow us to simplify this problem. For example, in aircraft navigational systems, the observation of the Earth’s surface relief and generation of the corresponding image are carried out by scanning the local transmission and receiving diagrams. Many of the other optical-electronic equipment in navigational systems have the same local transmission and receiving diagrams, for example, devices for observation. This allows us to change the object space when the depth of this space is not so high in value and the

Copyright 2005 by CRC Press

1598_C12.fm Page 503 Monday, October 11, 2004 3:22 PM

Theory of Space–Time Signal and Image Processing in Navigational Systems

503

object space is far from the optical receiver by plane, the brightness of which is deﬁned by the function of two coordinates x and y along the direction of the viewﬁnder and the time and wavelength (x, y, t, λ). In addition, the interaction of such optical-electronic equipment with the majority of the surface and sea objects is carried out in the course of short time periods. This allows us to consider that the brightness is not varied in time except in the case of fast-moving objects or objects possessing the ﬂuctuating brightness.42 The distribution function of brightness within the area limits of the observed object can have various forms. Thus, we can simplify the problem of deﬁnition of the distribution function of brightness due to speciﬁc conditions of employment of the pattern recognition optical-electronic systems. One of the peculiarities is that the procedure of object detection starts for distances that are larger in comparison with the dimensions of the object. For this reason, the optical system must be photosensitive. This implies that aberrations of the optical equipment cannot be reduced to negligible values taking into consideration the ﬁnite dimension of the vision ﬁeld and the bandwidth of the spectral range. Similar optical systems cannot distinguish the components of the shape of the highly remote objects and considers these objects in the form of the point sources. Therefore, the total dimensions of the object’s shape play a secondary role. In this case, the energy characteristic Jen is characterized as a power Pen radiated by the point source or by the source, the dimensions of which are negligible in comparison with the distance between the radiation source and the receiver or detector of the navigational system, in the solid angle Ω. Based on these statements, we can write J en =

dPen . dΩ

(12.117)

When the receiver and the radiation source are approaching, the condition that the angle dimensions of the object be negligible is not satisﬁed and the representation of the object in the form of the point radiation source leads to high-level errors. For this reason, we use the statement of the optical power. Let us consider brieﬂy the main characteristics and peculiarities of the background radiation sources. In the consideration of the object detection problem, we can assume that the natural radiation sources limited by the radar range and noise immunity of the optical-electronic navigational systems are the background noise. For this reason, the background noise can be the atmosphere, the clouds, the Earth’s surface, the sea surface, the stars in sky, etc. During navigation operations, some of the background radiation sources can serve as landmarks. In that case, it is necessary to solve the object detection problem using other background noise radiation sources.43,44

Copyright 2005 by CRC Press

1598_C12.fm Page 504 Monday, October 11, 2004 3:22 PM

504

Signal and Image Processing in Navigational Systems

Summing the background radiation noise, as well as object radiation, is deﬁned by two components: (1) the internal radiation and (2) radiation scattered by the sun and other external sources. The characteristics of the background noise sources, as noted previously, have more variety compared to the characteristics of the detecting objects. This phenomenon can be explained by a set of features of the background noise source formation. First, there is a variety of conﬁgurations and their omponents, which have various physical natures. The dimensions of the background noise sources are various but, as a rule, are greater than the dimensions of the object. Therefore, it is necessary to consider the components of the background noise sources in the three-dimensional space, for example, in the analysis of cloud forms. The essential differences arise in the deﬁnition of the characteristics of the background noise sources because of the unpredicted appearance and interchange of various gradations of the background noise sources and their elements under scanning by optical–electronic navigational systems. It should be noted that the radiation characteristics of the background noise sources depend highly on external conditions, i.e., the weather, the season, rain, snow, wind, etc. All these factors allow us to consider the signal as a stochastic process. The total deﬁnition of the three-dimensional stochastic ﬁeld structure of the background noise sources leads to the multidimensional distribution laws of brightness, which is a very complex problem. In practice, the problem of deﬁning the statistical characteristics of the stochastic ﬁeld of brightness is reduced to more simple interpretations. For example, we often use the one-dimensional probability distribution densities, correlation functions, and power spectral densities to analyze the stochastic ﬁelds in the form of underlying surfaces or the Earth’s surface relief. We can use these statistical characteristics in the deﬁnition both of the one-dimensional and of the twodimensional stochastic ﬁelds at their stationary state and isotropy conditions. For this purpose, it is necessary to estimate these statistical characteristics for the given Earth’s surface relief. This estimation allows us to deﬁne the degree of error and the applicability of the estimation in solving the speciﬁc problem. Moreover, we can consider the probability distribution density of brightness as a stationary process inside the individual characteristic zones, in which the total surface of the object can be divided into parts. The character of each zone is deﬁned, for example, by the homogeneity of the underlying surface. Sometimes it is possible to obtain the theoretical models of the background noise sources both with nonrandom and random parameters.45 However, a great amount of data can be obtained by experimental investigations of the main radiation characteristics, such as the coefﬁcient of reﬂection, coefﬁcient of brightness, and radiant emittance. As the main characteristic of radiation of the background noise sources we use the parameters ελ and µλ. The values of these coefﬁcients can be considered as the basis for the diagram or graph of the background noise source regions in the coordinate system of primary features of the pattern recognition problem or space of states, and, in the ﬁrst approximation, the Copyright 2005 by CRC Press

1598_C12.fm Page 505 Monday, October 11, 2004 3:22 PM

Theory of Space–Time Signal and Image Processing in Navigational Systems

505

coordinates. In the last case, two variants caused by the totality of the representations of the coefﬁcients ελ and µλ can arise. As an example of the ﬁrst variant, we can consider the case when statistical characteristics such as the mean and variance of the coefﬁcients ελ and µλ are given. Then the problem of the graph representation of the background noise sources is solved in the simplest way if the a priori probability distribution density of the coefﬁcients ελ and µλ is known. The problem of the graph representation of the background noise sources for the second case is more complex, especially if there is a deﬁcit of statistical data. At the present time, this case is more typical because, in spite of the large amount of references we are not able to obtain a completely satisfactory statistical picture of radiation and scattering for the background noise sources. This circumstance forces us to use a set of techniques and tools that will allow us to reach sequentially a satisfactory approximation of the complete deﬁnition. As one of these simplest procedures, we can use the technique of sequential or systematic approximation with the attraction of additional information about variations of chosen features at each step. The geometric representation of this technique using the mean of the spectral coefﬁcients of radiation and reﬂection as initial data is very evident. The mean of the coefﬁcients of radiation and reﬂection are used as the coordinate axes. We can deﬁne the regions corresponding to various forms of the background noise source images using these coordinates. The ﬁrst step is completed using this procedure. During the second step, an extension of the deﬁnition of these regions caused by the attraction of additional information data is carried out. These additional information data would be the results of experimental investigations obtained by the angle characteristics of radiation discussed in many references. The third step is devoted to a more precise deﬁnition of the region of the given form of the background noise sources obtained by the knowledge of brightness conditions, etc. We repeat this operation until we will have sufﬁcient information about all parameters inﬂuencing the variations of the coefﬁcients ελ and µλ. Obviously, the regions obtained this way cannot be considered as the model of features of the investigated object image. However, there is a possibility to correct them using further procedures steps obtained from receiving new and more precise information. In view of the fact that there is a relationship between the statistical characteristics and physico-geographical essence of the natural Earth’s surface relief, it is worthwhile to consider the seasonal background noise source regions for the various regions of the Earth’s surface relief. The environment in which optical–electronic navigational systems operate greatly stimulates the statistical characteristics of the optical signals. During propagation, optical signals have a great stimulus of the environment in comparison with the propagation of radio signals. In the general case, three main phenomena deﬁne the laws of propagation of optical signals:46,47 absorption, scattering, and turbulence. The ﬁrst and second phenomena deﬁne the average fading of the electromagnetic ﬁeld under ﬁxed atmospheric conditions and Copyright 2005 by CRC Press

1598_C12.fm Page 506 Monday, October 11, 2004 3:22 PM

506

Signal and Image Processing in Navigational Systems

comparatively slow variations of the electromagnetic ﬁeld characteristics under changed meteorological conditions. The third phenomenon is the turbulence that causes rapid variations of the electromagnetic ﬁeld characteristics observed under any meteorological conditions. Moreover, the structure of the received optical signals can be signiﬁcantly changed compared to the original signal, which because of turbulence, generates a multibeam effect. The ratio between the radiation past the atmospheric layer with the thickness equal to l and the incident radiation is characterized by the coefﬁcient of attenuation ~ β depending on the wavelength, in the general case. Thus, based on the preceding statements, we can consider that β˜ = β ab + β sc + β tub ,

(12.118)

where βab is the coefﬁcient of molecular absorption, βsc is the coefﬁcient of scattering by particles, and βtub is the coefﬁcient of scattering by nonhomogeneities caused by turbulence. The value ϖ 0 = β˜ l is called the optical thick˜ ness. The value ϖ = e −βl = e − ϖ0 is called the coefﬁcient of transparency. There is also the effect of back scattering, which makes the operating characteristics of navigational systems poor. If the source radiation is concentrated within the limits of the deﬁnite solid angle, the receiver processes both the radiation reﬂected from the surface and the source radiation scattered by particles being within the limits of the solid angle. This scattered radiation with the high concentration of aerosol particles having various dimensions within the limits of the viewﬁnder of the navigational system receiver can create a very high level of background noise and the detection of landmarks will be very difﬁcult.48,49 The mathematical model of radiation attenuation caused by the phenomena discussed in the preceding text is very complex and is based on quantum mechanics mathematics. Therefore, we use approximated methods or results of experimental investigations during the analysis. Similar techniques are discussed in References 45 and 50.45,50 Optical receivers or detectors employed in aircraft navigational systems, in their principal schemes and constructions, are of a great variety. The general features of all optical receivers and detectors are image-forming, ampliﬁcation of brightness at the input eye of the optical receiver or detector, and radiation ﬁltering by energy spectrum. Moreover, optical receivers and detectors can carry out other functions, for example, scanning, separation or summation of the radiation stream, ﬁltering by polarization power, ensurance of variables increasing with various view ﬁelds, etc. The view ﬁeld of optical receivers and detectors, as a rule, is not so large so that the angle coordinates of individual elements of the image are proportional to the linear dimensions in planes that are orthogonal to the optical axis of the receiver or detector. Therefore, the measurement of the linear position of the observed object is equivalent to bearing. As was noted in the preceding text, object radiation can be coherent or noncoherent. Therefore,

Copyright 2005 by CRC Press

1598_C12.fm Page 507 Monday, October 11, 2004 3:22 PM

Theory of Space–Time Signal and Image Processing in Navigational Systems

507

we often call the signal processing in the optical receiver or detector as coherent or noncoherent signal processing. In noncoherent radiation, the distribution of lightness E(x, λx, y, λy) is formed in the image space in which the lightness is proportional to the brightness (x, y) of associated points in the object space. If the object radiation is coherent, the optical receiver or detector is able to fulﬁll the transformation operation of the complex amplitudes E(x, y) in the two-dimensional space-frequency Fourier spectrum. These transformations are accompanied by losses in the energy of the signal and distortions of the signal due to aberrations of the optical receiver or detector. If these distortions of the signal are absent, i.e., the optical receiver or detector is the ideal system, then, in the large distances between the optical receiver or detector and the observed object, the plane of the optical receiver or detector forming the moving image coincides with its back focal plane. Thus, each point of the observed object with the coordinates λx and λy is represented in the focal plane by the point with the coordinates λx′ and λy′. The lightness corresponding to the information signal forming in the moving image in this way is given by E( x − λ x , y − λ y ) = ϖop π ( x − λ x , y − λ y )sin 2 ϑ ,

(12.119)

where (x, y) is the brightness contrast of the information signal in the moving image, which is deﬁned in the plane of the input eye of the optical receiver or detector; ϖop is the coefﬁcient of transparency of the optical receiver or detector; and ϑ is the back aperture angle. The no-ideal optical receiver or detector possesses such phenomena as diffracting scattering and abberation. These are the reasons for a fuzziﬁed image forming in the plane of the object observation. The fuzziﬁed image is characterized by the function of scattering (x, λx, y, λy), the physical form of which is that this function is the radiation at the point (λx′, λy′ ) when the current equal to unity is directed to the point (x′, y′). This deﬁnition explains the normalization of the function of scattering ∞

∫ (x, λ , y, λ ) dx dy = 1. x

y

(12.120)

−∞

The condition of normalization is the following. The fuzziﬁed stream at the point (λx, λy) must be equal to the initial incident radiation stream. After normalization, the function of scattering is often called the weight function. The function of scattering is related to the couple of the Fourier transform with the aperture function deﬁning the view ﬁeld of the optical receiver or detector. In the inﬁnite diameter of the optical lenses of the receiver or detector, the aperture function takes the square waveform shape, the base of which tends to approach inﬁnity. Therefore, for this optical lens, the function of scattering is deﬁned by the delta function that corresponds to Copyright 2005 by CRC Press

1598_C12.fm Page 508 Monday, October 11, 2004 3:22 PM

508

Signal and Image Processing in Navigational Systems

the ideal optical receiver or detector. For the no ideal optical receiver or detector, the total lightness at the point (x′, y′) at the object image plane will be equal to sum of lightnesses in terms of the stream scattering along the direction to all elements (dx′, dy′). If the fuzziﬁed images are the same at all points of the view ﬁeld of the optical receiver or detector, we can write ∞ ∞

E( x , y) =

∫ ∫ E(λ

x

− x , λ y − y) ⋅ (λ x , λ y ) dx dy.

(12.121)

−∞ −∞

For all cases of optical observations, navigational systems use optical receivers or detectors having responses by the energy of the incoming signals, i.e., optical receivers or detectors are able to ﬁx the radiant contrast of the observed object. This statement is caused by the wideband signals used and by the uncertainty in phase of their frequency components.51 The optical receivers or detectors operating by the radiant contrast or brightness contrast can be divided into the following classes: integral optical receivers or detectors, optical receivers or detectors with sequential searching, and multielement optical receivers or detectors. In the ﬁrst, the total energy of the incoming signals changes the parameters of the optical receiver or detector. An example of such receivers or detectors is the photo resistor, the resistance of which changes proportionally to the incident light stream because of the internal photo effect. Optical receivers or detectors of the second class are used for sequential in-time construction of the object image in the observation plane. Transformation of the two-dimensional object image into the one-dimensional electric signal is carried out by the sweep in time on the observation plane. Examples of optical receivers or detectors of this kind are TV transmitters with multiframe sweep and heat vision receiver that allow us to construct the observed object image in the infrared optical range. The multielement optical receiver or detector functioning is based on simultaneous representation of all radiating elements of the observed object in the plane in which the object image is formed. For these optical receivers or detectors, we use the matrix structure. The matrix receivers or detectors are the totality of elements spaced into the same plane that are sensitive to the incident radiation stream. In other words, the matrix receivers or detectors are constructed based on a set of the integral optical receivers or detectors with individual outputs. The output signals of the optical receivers or detectors correspond by level to that part of the total radiation energy which covers an area occupied by the optical receivers or detectors. An example of these optical receivers or detectors is the matrix receiver or detector consisting of photo elements. In the majority of cases, we use the integral and spectral sensitivities by the current and by the voltage ε(x′, y′). Thus, the main characteristic of the optical receiver or detector is the ratio between the energy characteristic of

Copyright 2005 by CRC Press

1598_C12.fm Page 509 Monday, October 11, 2004 3:22 PM

Theory of Space–Time Signal and Image Processing in Navigational Systems

509

the signal at the receiver or detector output and the energy characteristic of the radiation stream causing this output signal. The output signal of the optical receiver or detector can be written in the following form: ∞ ∞

=

∫ ∫ ε(x′, y′) ⋅ E(x′ − λ′ , y′ − λ′ ) dx′dy′. x

y

(12.122)

−∞ −∞

The formula in Equation (12.122) is true under the following conditions: the optical receiver or detector is inertialess and the variable is the total response of the individual element responses. Otherwise, the formula in Equation (12.122) is cumbersome.

12.6 Peculiarities of the Formation of the Earth’s Surface Radar Image To estimate the operating ability of the radar navigational system, taking into consideration features of the landmarks and model images, the preliminary generation of radar maps is widely used. However, the preliminary radar image obtained with the use of radar ﬁeld models is considered a cheaper operational process. The following initial premises operate in the construction of radar images of the Earth’s surface based on experimental study: the amplitude of the target return signals from the Earth’s surface relief, for example, the steppe, the forest, the arable land, and so on, obeys the Rayleigh probability distribution density, and the phase of these target return signals is distributed uniformly within the limits of the interval [0, 2π]; the mean and variance of the target return signal amplitude are deﬁned by the speciﬁc effective scattering area S° of the background noise source; S° is deﬁned by the Earth’s surface relief form, such as the steppe, the forest, the arable land, the ﬁelds, the city districts, etc. The background noise source distribution S°(x, y) is deﬁned by the speciﬁc relief observed by the radar navigational system. There are powerful individual pulsed target return signals reﬂected from the bridges, the electric power transmission lines, etc., the amplitudes of which obey the Rayleigh probability distribution density. The amplitude of the target return signal from the background noise sources can be expressed as two cofactors: S1 ( x , y) = A( x , y) ⋅ S° ,

(12.123)

where S° corresponds to the Earth’s surface relief background (see Table 12.3). The dimensions of the Earth’s surface relief background regions with

Copyright 2005 by CRC Press

1598_C12.fm Page 510 Monday, October 11, 2004 3:22 PM

510

Signal and Image Processing in Navigational Systems TABLE 12.3 Values of the Speciﬁc Effective Scattering Area S°, dB

Background

–40…+45 –30 –20 –15 –10…+10

Water Concrete Steppe Forest City

the same value of S° are greater in comparison with the dimensions of the resolution elements of the output radar image. The spectral bandwidth of spatial frequencies of the background noise sources and the correlation function are deﬁned by the Earth’s surface relief types. The cofactor A(x, y) is a stationary stochastic process with the variance σA2 and obeys the Rayleigh probability distribution density52 2

f ( a) =

2 A − σAA ⋅e . σ 2A

(12.124)

The amplitudes of the target return signals from the point contrast landmarks can be presented, as a rule, by the delta function N

∑

S2 ( x , y) = A( x , y)

Sk ⋅ δ( x − xk , y − yk ) ,

(12.125)

k =1

where Sk is the speciﬁc effective scattering area of the k-th landmark with the coordinates xk and yk. The value of Sk is deﬁned by the landmark type (see Table 12.4). If the point landmarks are very close to each other, they form a lengthy image with deﬁnite conﬁguration, for example, bridge, road, etc. Thus, the total electromagnetic ﬁeld strength distribution, without taking into account the directional diagram of the navigational system, can be written in the following form: TABLE 12.4 Kinds of Landmarks S k, m 2 1…10 3…5 15…20 150 14 · 103

Copyright 2005 by CRC Press

Object Car Small aircraft Large aircraft Small ship Tanker

1598_C12.fm Page 511 Monday, October 11, 2004 3:22 PM

Theory of Space–Time Signal and Image Processing in Navigational Systems

[

S( x , y) = S1 ( x , y) + S2 ( x , y) = A( x , y)

N

S°( x , y) +

∑

511

]

Sk ⋅ δ( x − xk , y − yk ) .

k =1

(12.126) The generalized detector for the radio signal with the envelope has the same structure as in the case of the generalized receiver detecting the same envelope in the background additive wideband noise. Therefore, the power spectral density is doubled in comparison with the power spectral density of the additive high-frequency noise. Based on this we can consider Equation (12.126) as the basic formula in the generation of the radar image. The model of the radar navigational system forming the Earth’s surface relief image can be presented in the form of the connected sequential receiver–transmitter channel, the signal processing system, and the decision-making device (see Figure 12.16). The receiver–transmitter channel can be presented in the form of the linear system with the pulse response g(x, y) with respect to the amplitude envelope of the signals. In scanning the Earth’s surface relief region, the amplitude of the generated signal can be determined in the following form: N

(λ x , λ y ) = cˆ

∑ A(x , y ) ⋅ k

k

Sk ⋅ g( x − λ x − xk , y − λ y − yk ) + n( x , y),

k =1

(12.127) where λx and λy are the coordinates of the image on the observation plane XOY, and cˆ is the transmission coefﬁcient of the radar channel. The function g(x, y) can be expressed as the product of the two functions g(x) and g(y). The ﬁrst function g(x) is deﬁned by the width θ0 of the directional diagram and sloped plain. The second function g(y) is deﬁned by the duration τp of the pulsed searching signal. In side scanning (see Figure 12.17), the axis x is the course distance, the axis y is the side distance, and the directional diagram can be approximated by the Gaussian law n S (x, y )

ReceiverTransmitter

+ U(λx , λy)

FIGURE 12.16 Radar channel model.

Copyright 2005 by CRC Press

UM (λx , λy) Signal Processing

DecisionMaking System

1598_C12.fm Page 512 Monday, October 11, 2004 3:22 PM

512

Signal and Image Processing in Navigational Systems y

y1 y2

θ0

x

FIGURE 12.17 Side scanning.

g( x , y ) = e

− ax x 2 − ay y 2

, ax =

c1 , y 2θ0

and

ay =

c2 , τ 2p

(12.128)

where c1 and c2 are the constant coefﬁcients. Thus, the resulting image is the totality of bright ﬂashes on the indicator observed in the background noise. The dimensions of each ﬂash are deﬁned by the coefﬁcients ax and ay , i.e., by the width θ0 of the directional diagram and the duration τp of the pulsed searching signal. The information signal in the moving image during M periods of scanning in accordance with Equation (12.127) takes the following form: M

ˆ M (λ x , λ y ) = cM

∑ A(x , y ) ⋅ S ⋅ e k

k

−2 ax ( x − λ x − xk )2 − 2 ay ( y − λ y − yk )2

k

.

(12.129)

k =1

In this case, the criterial function has the following form: I = R(0) ⋅ e

− ax ∆xk2 − ay ∆yk2

,

(12.130)

where N

R(0) = cˆ 2 M 2

N

∑∑

Sk Sl ⋅ A( xk , xl ), ∆xk = xk − xk* , and ∆yk = yk − yk* .

k =1 l=1

(12.131) Copyright 2005 by CRC Press

1598_C12.fm Page 513 Monday, October 11, 2004 3:22 PM

Theory of Space–Time Signal and Image Processing in Navigational Systems

513

The discrimination characteristics in measuring the parameters λx and λy are determined by ∂I − a ∆x 2 − a ∆y 2 = 2 R(0)ax ∆xk ⋅ e x k y k , * ∂λ x

(12.132)

∂I − a ∆x 2 − a ∆y 2 = 2 R(0)ay ∆x y ⋅ e x k y k . * ∂λ y

(12.133)

The formulae in Equation (12.131)–Equation (12.133) are written without taking into account the background noise. Thus, in the observation of the point landmarks, the shape of the directional diagram and the duration of the pulsed searching signal play the main role in the process of forming the discrimination characteristic. In practice, this dependence is very complex. The phenomenon is explained by the following factors: the high-level brightness contrast landmark cannot be deﬁned by the delta function; the linear model of the radar channel has an approximated character; the background noise source leads to the fuzziﬁed image of the observed landmark totality; and the characteristics of the images of the same Earth’s surface relief hardly depend on weather conditions, seasonal conditions, or observation direction. There are some other techniques of forming the Earth’s surface image using a topographical relief map. The topographical map allows us to deﬁne the area occupied by cities, forests, steppes, etc. The total energy reﬂected by the area is limited by the values of 0.5cτp , where c is the velocity of electromagnetic radiation and of the width of the horizontal-coverage directional diagram. This total energy is the sum of the energies caused by area elements. In forming the images, we use the coordinate lattice with the distance cτ between the horizontal lines equal to 2sinp ϕ , where ϕ is the angle of scanning. A distance between the vertical lines is deﬁned by the shape of the horizontal-coverage directional diagram.

12.7 Foundations of Digital Image Processing The wide use of matrix receivers or detectors generates the application of digital signal processing. Digital signal processing methods are widely used to determine the criterial functions in navigational systems for the continuous signals also. Two procedures form the basis of digital signal processing: (1) the observed stochastic space–time process sampling and (2) the quantization of the observed stochastic space–time process.

Copyright 2005 by CRC Press

1598_C12.fm Page 514 Monday, October 11, 2004 3:22 PM

514

Signal and Image Processing in Navigational Systems G

y

x FIGURE 12.18 The lattice function.

Sampling is a representation of the stochastic space–time process in the form of the totality of readings corresponding to the chosen discrete values of the arguments (x1, x2, …, xk; y1, y2, …, yl). The values of the arguments (xk, yl) are the chosen multiples to intervals ∆x and ∆y, respectively, which are called sampling intervals. If we consider the space–time signal a(x, y, t), the discrete readings with respect to the variables x and y are carried out for the ﬁxed instant of time. In the discrete representation of a stochastic process, it is very convenient to use the lattice function (see Figure 12.18): ∞

∞

∑∑

∞

alat (i∆x , j∆y) =

i=−∞ j=−∞

∞

∞

∑ ∑ ∫ a(x, y) ⋅ δ(i∆x − x) ⋅ δ((j∆y − y) dx dy.

i=−∞ j=−∞ −∞

(12.134) As follows from the interpolation theory of continuous spatial signals, the only realization a(x, y) can be constructed for the use of the given lattice function. The correlation lengths ∆x0 and ∆y0 of the realization a(x, y) are satisﬁed by the following condition: ∆x ≤ ∆x0

and ∆y ≤ ∆y 0 .

(12.135)

In other words, the continuous stochastic image can be presented in a discrete form always, i.e., in the form of the lattice function with discrete intervals for the condition given by Equation (12.135). Therefore, the continuous image a(x, y) can be reconstructed using the readings of the discrete image by the following interpolation:18 ∞

a( x , y) =

∞

∑∑a

lat

i = −∞ j = −∞

Copyright 2005 by CRC Press

(i∆x , j∆y) ⋅ f ( x − i∆x , y − j∆y),

(12.136)

1598_C12.fm Page 515 Monday, October 11, 2004 3:22 PM

Theory of Space–Time Signal and Image Processing in Navigational Systems

515

where f(x, y) is the deterministic interpolation function. We can assume that the higher interpolation certainty, i.e., the reconstructed ﬁeld and the initial image are equivalent in the statistical sense, can be obtained in the case if the interpolation function f(x, y) coincides with the coefﬁcient of the space correlation function of the original signal. If the discrete sequence alat(i∆x, j∆y) comes in at the input of the spatial ﬁlter with the pulse characteristic f(x, y), the ﬁlter response corresponds to the initial image a(x, y). The correspondence is more precise, the closer the function f(x, y) is to the coefﬁcient of spatial correlation of this image. The most convenient and widely used discrete representation of the continuous image is the presentation in the form of the step envelope lattice function53 (see Figure 12.19): a(i , j ) = a( x , y ) ⋅ f1 ( x , y ),

(12.137)

where K

f1 ( x , y ) =

M

∑ ∑ p(x − i∆x, y − j∆y)

(12.138)

i=1 j=1

is a totality of the K × L identical square waveform normalized pulses ∞ ∞

∫ ∫ p(x, y) dx dy = 1 .

(12.139)

−∞ −∞

G

x FIGURE 12.19 The step envelope of the lattice function.

Copyright 2005 by CRC Press

1598_C12.fm Page 516 Monday, October 11, 2004 3:22 PM

516

Signal and Image Processing in Navigational Systems

Up till now, we discussed the quantization of the image a(x, y) without taking into consideration the spatial noise n(x, y). Let us consider the moving image in the following form: X( x , y ) = a( x , y ) + n( x , y ).

(12.140)

We assume that the correlation length of the noise is much less than that of the signal. In this case, the use of the condition given by Equation (12.135) can give rise to additional distortions in the reconstruction of the image. The essence of these distortions is the following. Let the information signal a(x) with the correlation length ∆x0 be observed in the presence of additive noise with the correlation length ∆xn << ∆x0 (see Figure 12.20). If the sampling interval is equal to ∆xn << ∆x0 , we can lose some information (see Figure 12.21) in the reconstruction of the initial moving image. To avoid these losses we should carry out ﬁltering for the moving image before sampling. Here, the correlation length of the noise is close to the correlation length of the information signal, and additional distortions of the information signal a(x) are absent. In practice, the space–time signal a(x, y, t) at the ﬁxed instant of time deﬁned by the lattice function alat(i∆x, j∆y), i = 1, 2, …, K, j = 1, 2, …, L can be determined in the form of the step envelope A(i, j) or in the form of the rectangular matrix of readings with the number of elements equal to K × L. The criterial function can be determined in the following form: K

I (m, n) = cˆ

L

∑ ∑ A(i, j) ⋅ X(i − m, j − n), i=1 j=1

X (x, y)

a(x) + n(x) a(x)

∆x x 0 FIGURE 12.20 Additive mixture of signal and noise.

Copyright 2005 by CRC Press

(12.141)

1598_C12.fm Page 517 Monday, October 11, 2004 3:22 PM

Theory of Space–Time Signal and Image Processing in Navigational Systems

517

X (x, y)

a* (x) a(x)

∆x x 0 FIGURE 12.21 Quantization of signal and noise. 1 . The criterial function I(m, n) given by Equation (12.141) is the where cˆ = KL ﬁxed moment of the second order for the digital image. The moments of higher order can be determined in the following form: K

µ pq = cˆ

L

∑ ∑ A (i, j) ⋅ X (i, j). p

q

(12.142)

i=1 j=1

In the case when the means and <X(i, j)> are not equal to zero, the central moments are given by K

µ 0pq = cˆ

L

∑ ∑ [A(i, j) − < A(i, j) >] ⋅ [X(i, j) − < X(i, j) >] , p

q

(12.143)

i=1 j=1

1 where cˆ = KL , K = ∆Xx , L = ∆Yy . The discrete power spectral density of the stationary two-dimensional stochastic ﬁeld A(i, j) takes the following form:

S(ω x , ω y ) =

1 MN

M

N

∑ ∑ I(m, n) ⋅ e

− j(

mω x M

+

nω y N

)

.

(12.144)

m=1 n=1

We can represent the continuous signal in the discrete form in the spectral range. For this purpose, the totality of the delta functions used in forming the lattice functions can be determined by the Fourier-series expansion ∞

f 2 ( x , y) =

∞

∑∑

k1 = − ∞ k2 = − ∞

∞

δ( x − k1 ∆x , y − k2 ∆l) =

∞

∑ ∑ B(k , k ) ⋅ e 1

2

j( k1 2 π x + k2 2 π y ) ∆x

∆y

,

k1 = − ∞ k2 = − ∞

(12.145) Copyright 2005 by CRC Press

1598_C12.fm Page 518 Monday, October 11, 2004 3:22 PM

518

Signal and Image Processing in Navigational Systems

where B(k1, k2) are the coefﬁcients of the Fourier-series expansion: 1 ∆x∆y

B( k1 , k2 ) =

0.5 ∆x 0.5 ∆y

∫ ∫

f 2 ( x , y) ⋅ e

− j( k1 2 π x + k2 2 π y ) ∆x

∆y

(12.146)

dx dy.

− 0.5 ∆x − 0.5 ∆y

The delta function has a ﬁltering feature. For this reason, the coefﬁcients B(k1, k2) are constant values and given by B(k1, k2) = ∆x1∆y . Consequently, we can write ∞

∞

∑∑

1 f 2 ( x , y) = ∆x∆y k

e

j( k1 2 π x + k2 2 π y ) ∆x

∆y

(12.147)

.

1 = − ∞ k2 = − ∞

The power spectral density for this function takes the following form:

S(ω x , ω y ) =

∞

∞ ∞

∞

∑ ∑ ∫ ∫e

1 ∆x∆y k

1 = − ∞ k2 = − ∞

×e

− j(ω x x + ω y y )

dx dy =

j( k1 2 π x + k2 2 π y ) ∆x

∆y

−∞ −∞ ∞

∞

∑ ∑ δ(ω

4π 2 ∆x∆y k

x

− k1

2π ∆x

, ω y − k2

2π ∆y

).

1 = − ∞ k2 = − ∞

(12.148) The power spectral density of the digital image can be deﬁned using the power spectral density of the continuous image: S(ω x , ω y ) =

4π 2 ∆x∆y ∞

×

∞

∞ ∞

∑ ∑ ∫ ∫ S(ξ, η) ⋅ δ(ξ − ω

x

− k1

2π ∆x

, η − ω y − k2

2π ∆y

) dξ d η

k1 = − ∞ k2 = − ∞ − ∞ − ∞

=

∞

∞

∑ ∑ S(ω

4π 2 ∆x∆y k

x

− k1

2π ∆x

, ω y − k2

2π ∆y

).

1 = − ∞ k2 = − ∞

(12.149) Let us assume that the space–time signal a(x, y, t) obeys the probability distribution density f(a) in the range of the possible values Ω. Quantization assumes the partition of the range Ω for the p quantization intervals Ωp. In doing so, all values a(x, y) within the limits of the interval Ωp have the same estimation value ap. The parameter p is called the quantization volume. The Copyright 2005 by CRC Press

1598_C12.fm Page 519 Monday, October 11, 2004 3:22 PM

Theory of Space–Time Signal and Image Processing in Navigational Systems

519

parameter ap is called the quantization level. The error of quantization of the signal a(x, y) can be determined in the following form: P

ε2 =

∑ ∫ (a − a ) ⋅ f (a) da . 2

p

(12.150)

p = 1 Ωp

If the number of quantization levels is high, the probability distribution density of the quantized signal can be thought as constant. In this case, we can write P

ε2 =

∑ p=1

∫

P

∑ f (a )[(d

f ( ap ) ( a − ap )2 da ≈ 0.33 Ωp

p

p +1

− ap )3 − (dp − ap )3 ],

p=1

(12.151) where dp and dp+1 deﬁne the p-th quantization interval Ωp. It is well known that the optimal level of quantization is determined by ap = 0.5(dp+1 + dp). In other words, if the quantization level ap is at the middle of the quantization interval, the value ε is minimum.

12.8 Conclusions The totality of landmarks used in solving navigational problems such as the Earth’s surface during navigation of the aircraft ﬂight track, the image of starry sky, etc., is observed, as a rule, in the background of interferences and noise generated by various sources. For this reason, the effectiveness of solving the navigational problems is deﬁned by the quality of the components used in obtaining the moving image, i.e., by the quality of signal and image processing of information components of the moving image containing useful information regarding the totality of landmarks. It is obvious that the methods and techniques of space–time signal and image processing in navigational systems are deﬁned by the nature of the used signals. The moving image of the Earth’s surface relief is formed due to periodic changes in the aircraft ﬂight altitude. If we solve the navigational problem of the moving ship, the sea depth is measured. In both these cases the moving image is a function of time due to the moving radar navigational system. Here, the signal and image processing is similar to that in time just as in the observation of the geophysical ﬁeld. Thus, methods and techniques of spatial target return signal can be classiﬁed into space signal and image processing, space–time signal and image processing, and signal and image processing in time.

Copyright 2005 by CRC Press

1598_C12.fm Page 520 Monday, October 11, 2004 3:22 PM

520

Signal and Image Processing in Navigational Systems

The solution of the navigational problems, i.e., the deﬁnition of location of the navigational object, is carried out under ambiguous conditions. This uncertainty is formed by the following main factors: external interference, sensor noise, interference that can arise with the moving radar of navigational system, indeﬁnite information regarding landmark coordinates, etc. Incorrect knowledge of location of landmarks, i.e., the relative position of the navigational system in the coordinate system, is referred to as a priori uncertainty in the deﬁnition of the coordinates of the navigational object. For reasons already noted, we can discuss only the estimation of some coordinates of the navigational object that can be obtained anyway. In the case when the estimation of the coordinates of the navigational system is deﬁnitely the best, we can consider that this estimation is optimal. Thus, the main problem in navigational systems is the deﬁnition of the optimal estimation of the navigational object coordinates, for example, the deﬁnition of the true aircraft ﬂight track, using information in the signals of various physical sources. The results discussed in this chapter indicate that we have a good chance to use the generalized detector in navigational systems in the signal processing of the target return signal with stochastic parameters, and that the estimation of the deﬁnition of the navigational object coordinates has high accuracy. We proceed as follows: construct the amplitude and phase tracking systems that control the appropriate parameters of the model signal generated at the output of the model signal generator of the generalized detector. In principle, the construction of the amplitude and phase tracking systems is made possible by using the condition T >> τc . The space–time signal and image processing by the use of the generalized approach to signal processing in the presence of noise is to deﬁne the maximum of the integral given by Equation (12.68). If the dimensions of the observed moving image are much more than the correlation length of the signal and the signal is the stationary stochastic process, the integral in Equation (12.68) matches the maximum of the likelihood functional. The deﬁnition of this maximum is carried out by the use of the generalized detector tracking systems discussed in Section 12.2. Thus, the discrimination characteristics are deﬁned along the corresponding axes Ix and Iy . Navigational systems that realize the generalized approach to signal processing in noise are called correlation-extremal systems because Equation (12.68) allows us to deﬁne the maximum of the likelihood functional only under the main conditions of the generalized detector functioning. The functions given by Equation (12.68) and Equation (12.69) are often called criterial functions. The criterial functions given by Equation (12.69) are proportional to the measurement errors because the background noise in the generalized approach to signal processing tends to approach zero in the statistical sense and the noise component 2

∫ ∫ a (x, y, Λ , t)n(x, y, t)dxdy of the correlation channel of the *

*

X Y

generalized detector, caused by the interaction between the model image

Copyright 2005 by CRC Press

1598_C12.fm Page 521 Monday, October 11, 2004 3:22 PM

Theory of Space–Time Signal and Image Processing in Navigational Systems

and noise, and the random component 2

521

∫ ∫ a(x, y, Λ, t)n(x, y, t)dxdy of the X

Y

autocorrelation channel of the generalized detector, caused by interaction between the moving image and noise, are compensated by each other in the statistical sense (see Chapter 11). Correlation extremal receivers or detectors constructed according to the generalized approach to signal processing in the presence of noise may be used in navigational systems without tracking systems. The use of the generalized receivers or detectors of this form implies the employment of the signals with known amplitude-phase-frequency structure. The comparison of the incoming moving image at the input of the navigational system with the model image deﬁned by the form of the signal and the a priori knowledge regarding the location of the signal on the observation plane allows us to estimate the true value of the parameter vector of Λ. Thus, an automatic control of this value of the parameter vector Λ is carried out. The main condition of the navigational system functioning without tracking devices is the correlation between the moving image X(x, y, t) and the model image a*(x, y, Λ*, t). In other words, we cannot recognize the information signal or image a(x, y, Λ, t) in the input moving image X(x, y, t). The use of the generalized detector with tracking systems (see Section 12.2) allows us to solve this problem, to recognize the type of the signal a(x, y, Λ, t), and to deﬁne and measure the parameter vector Λ. According to the character of the additive noise n(x, y) at the output of the navigational system, we can use the preliminary ﬁlter of the generalized detector of various forms. For simplicity, we assume that the noise n1(x, y) is the Gaussian narrowband process. Then, the criterial function for each estimation channel λx and λy is deﬁned by the derivative of the space–time correlation function of the signal derivatives with the weight coefﬁcients that are inversely proportional to the spectral bandwidth of the noise n1(x, y) in the region of spatial and ordinary frequencies. To realize this space–time signal in practice and the image processing algorithm based on the generalized approach to signal processing in the presence of noise to solve navigational problems, we should have a special memory block to store both the map of the ﬁeld (the model image) and the map of the ﬁrst partial derivatives of the map of the ﬁeld (the ﬁrst partial derivatives of the model image). In the observation of the spatial signals a(x, y) and space–time signals a(x, y, t), the pattern recognition principles are the following. We should ﬁnd a region on the observed plane XOY in which the function of intensity is similar to the predetermined function of intensity called the model function. In this case, the criterion of similarity is the quadratic measure given by Equation (12.91), which can be reduced to the correlation function between the moving and model images given by Equation (12.99) in the case of the Gaussian probability distribution density of the noise, where the parameters λx and λy deﬁne the shift between the model image relative to the moving image belonging to the same group of images. The maximum

Copyright 2005 by CRC Press

1598_C12.fm Page 522 Monday, October 11, 2004 3:22 PM

522

Signal and Image Processing in Navigational Systems

of the correlation function given by Equation (12.99) corresponds to the complete match between the model image and the moving image. The position of the model image in the coordinate system XOY is known a priori (see Figure 12.15). Thus, the deﬁnition of the maximum of the correlation function given by Equation (12.99) allows us both to solve the pattern recognition problem and to deﬁne and measure the position of the moving image in the coordinate system XOY. The composition of the p-channel generalized receiver allows us to solve the pattern recognition problem in the following manner. The moving image X(x, t) comes in at each input of the p-channel generalized receiver. The moving image X(x, t) is the discrete stochastic process. There is the discrete model image for each channel of the p-channel generalized receiver. The position of the model image at the initial instant of time is deﬁned by a priori knowledge and the estimation λi* obtained before. If for the i-th channel of the generalized detector we obtain the reduced value of σ ii2 ∗ , where σ ii2∗ = <(λ i − λ ∗i )2 >, we can make the decision a “yes” information signal in the moving image, and the information signal belongs to the same group as the model image. With the sequential pattern recognition technique, we can use the one-channel generalized detector with a set of model images. In this case, we should have p model images. This technique is used in a small sample of the moving image. If the sample of the moving image is large, the computing cost is very high. The environment in which the optical–electronic navigational systems operate acts as a great stimulus on the statistical characteristics of optical signals. During propagation, optical signals are greatly stimulated by the environment in comparison with the propagation of the radio signals. In the general case, three main phenomena deﬁne the laws of propagation of optical signals: absorption, scattering, and turbulence. The ﬁrst and second phenomena deﬁne the average fading of the electromagnetic ﬁeld in ﬁxed atmospheric conditions and comparatively slow variations of the electromagnetic ﬁeld characteristics under changed meteorological conditions. The third phenomenon is the turbulence that causes rapid variations of the electromagnetic ﬁeld characteristics observed under any meteorological conditions. Moreover, the structure of the received optical signals can be signiﬁcantly changed in comparison to the initial signal that, due to the turbulence, generates a multibeam effect. The ratio between the radiation past the atmospheric layer with the thickness equal to l and incident radiation is characterized by the coefﬁcient of attenuation β depending on the wavelength, in the general case. Optical receivers or detectors employed in aircraft navigational systems are of a great variety in terms of their principal schemes and constructions. The general features of all optical receivers and detectors are the forming of the image, ampliﬁcation of lightness at the input eye of the optical receiver or detector, and radiation ﬁltering by the energy spectrum. Moreover, optical receivers and detectors can perform other functions, for example, scanning,

Copyright 2005 by CRC Press

1598_C12.fm Page 523 Monday, October 11, 2004 3:22 PM

Theory of Space–Time Signal and Image Processing in Navigational Systems

523

separation or summation of the radiation stream, ﬁltering by polarization power, and ensuring the increase of the variable with various view ﬁelds, etc. Optical receivers or detectors operating by radiant contrast or brightness contrast can be divided into the following classes: integral optical receivers or detectors, optical receivers or detectors with sequential searching, and multielement optical receivers or detectors. In the ﬁrst class, the total energy of the incoming signals changes the parameters of the optical receiver or detector. An example of such receivers or detectors is the photo resistor, whose resistance changes proportionally to the incident light stream because of the internal photo effect. Optical receivers or detectors of the second class are used for the time sequential construction of the object image in the observation plane. The transformation of the two-dimensional object image into the one-dimensional electric signal is performed by the sweep in time on the observation plane. Examples of optical receivers and detectors in this case are the TV transmitter with multiframe sweep and the heat vision receiver, allowing us to construct the observed object image in the infrared optical range. The following are the initial premises in the construction of radar images of the Earth’s surface based on experimental study: the amplitude of the target return signals from the Earth’s surface relief, for example, the steppe, the forest, the arable land, etc., obeys the Rayleigh probability distribution density, and the phase of these target return signals is distributed uniformly within the limits of the interval [0, 2π]; the mean and variance of the target return signal amplitude are deﬁned by the speciﬁc effective scattering area S° of the background noise source; the speciﬁc effective scattering area S0 is deﬁned by the Earth’s surface relief form, such as the steppe, the forest, the arable land, the ﬁelds, the city districts, etc. The background noise source distribution S°(x, y) is deﬁned by the speciﬁc relief observed by the radar navigational system. There are powerful individual pulsed target return signals reﬂected from bridges, electric power transmission lines, etc., the amplitudes of which obey the Rayleigh probability distribution density.

References 1. Papoulis, A., Systems and Transforms with Application in Optics, Krieger, Malabar, FL, 1981. 2. Hansen, T. and Yaghjian, A., Plane-Wave Theory of Time-Domain Fields: Near-Field Scanning Applications, IEEE Press, Piscataway, NJ, 1999. 3. Felsen, L. and Marcuvitz, N., Radiation and Scattering of Waves, Prentice Hall, Englewood Cliffs, NJ, 1973. 4. Franceschetti, G., Iodice, A., Migliaccio, M., and Riccio, D., Scattering from natural rough surfaces modeled by fractional Brownian motion two-dimensional processes, IEEE Trans., Vol. AP-47, No. 9, 1999, pp. 1405–1415.

Copyright 2005 by CRC Press

1598_C12.fm Page 524 Monday, October 11, 2004 3:22 PM

524

Signal and Image Processing in Navigational Systems

5. Tuzlukov, V., A new approach to signal detection theory, Digital Signal Process. Rev. J., Vol. 8, No. 3, 1998, pp. 166–184. 6. Tuzlukov, V., Signal Processing in Noise: A New Methodology, IEC, Minsk, 1998. 7. Tuzlukov, V., Signal Detection Theory, Springer-Verlag, New York, 2001. 8. Tuzlukov, V., Signal Processing Noise, CRC Press, Boca Raton, FL, 2002. 9. Bochkarev, A., Optimal correlation navigational systems, A Broad Radio Electronics, No. 9, 1981, pp. 37–45 (in Russian). 10. Vorob’ev, V., Optical Location for Radio Engineers, Radio and Svyaz, Moscow, 1983 (in Russian). 11. Pavlov, Yu, Selevnev, A., and Tolstousov, G., Geoinformatic Systems, Mashinostroenie, Moscow, 1978 (in Russian). 12. Krasovsky, A., Beloglazov, I., and Chigin, G., Theory of Optimal Correlation Navigational Systems, Nauka, Moscow, 1979 (in Russian). 13. DeSanto, J. and Brown, G., Analytical techniques for multiple scattering from rough surfaces, in Progress in Optics XXIII, E. Wolf, Ed., North-Holland, Amsterdam, 1986. 14. Bossavit, A. and Mayergoyz, I., Edge elements for scattering problems, IEEE Trans., Vol. MG-25, No. 7, 1989, pp. 2816–2821. 15. Born, M. and Wolf, E., Principles of Optics, Pergamon, Oxford, 1980. 16. Thompson, A., Moran, J., and Swenson, G., Interferometry and Synthesis in Radio Astronomy, 2nd ed., John Wiley & Sons, New York, 2001. 17. Baldauf, J., Lee, S., Ling, H., and Chou, R., On physical optics for calculating scattering from coated bodies, J. Electromagnet. Waves Appl., Vol. 3, No. 8, 1989, pp. 725–746. 18. Baklitzky, V. and Yur’ev, A., Correlation Extremal Methods of Navigation, Radio and Svyaz, Moscow, 1982 (in Russian). 19. Goldsmith, P., Quasioptical Systems: Gaussian Beam Quasioptical Propagation and Applications, IEEE Press, New York, 1998. 20. Dogaru, T. and Carin, L., Time-domain sensing of targets buried under a rough air-ground interface, IEEE Trans., Vol. AP-46, No. 3, 1998, pp. 360–372. 21. Jiao, D. and Jin, J., Three-dimensional orthogonal vector basis functions for time-domain ﬁnite element solution of vector wave equations, IEEE Trans., Vol. AP-51, No. 1, 2003, pp. 59–66. 22. Baklitzky, V., The use of Kalman ﬁltering under synthesis of correlation-extremal systems, News of the USSR Universities. Radio Electronics, Vol. 25, No. 3, 1982, pp. 65–73 (in Russian). 23. Deans, S., The Random Transform and Some of Its Applications, Krieger, Malabar, FL, 1993. 24. Krasovsky, A., Optimal estimation in distributed systems deﬁned by the Green function, Automatics and Telemechanics, No. 10, 1981, pp. 25–32 (in Russian). 25. Capolino, F. and Felsen, L., Time-domain Green’s function for an inﬁnite sequentially excited periodic planar array of dipoles, IEEE Trans., Vol. AP-51, No. 2, 2003, pp. 160–170. 26. Felsen, L. and Capolino, F., Time-domain Green’s function for an inﬁnite sequentially excited periodic line array of dipoles, IEEE Trans., Vol. AP-48, No. 6, 2000, pp. 921–931. 27. Capolino, F. and Felsen, L., Frequency and time-domain Green’s function for a phased semi-inﬁnite periodic line array of dipoles, IEEE Trans., Vol. AP-50, No. 1, 2002, pp. 31–41.

Copyright 2005 by CRC Press

1598_C12.fm Page 525 Monday, October 11, 2004 3:22 PM

Theory of Space–Time Signal and Image Processing in Navigational Systems

525

28. Krasovsky, A., Field estimation under vector fuzzy measurements, Reports of Academy of the USSR. Cybernetics and Control Theory, Vol. 256, No. 5, 1981, pp. 387–393 (in Russian). 29. Ozaktas, H., Zalevsky, Z., and Kutay, M., The Functional Fourier Transform With Applications in Optics and Signal Processing, John Wiley & Sons, Chichester, U.K. 2001. 30. Zverev, V., Radio Optics, Soviet Radio, Moscow, 1975 (in Russian). 31. Zhao, L. and Cangellaris, A., GT-PML: Generalized theory of perfectly matched layers and its application to the reﬂectionless truncation of ﬁnite-difference time-domain grids, IEEE Trans., Vol. MTT-44, No. 12, 1996, pp. 2555–2563. 32. Jin, J., The Finite Element Method in Electromagnetics, John Wiley & Sons, New York, 1993. 33. Vishnevsky, V., Method of forming the standards that are invariant relative to shifts and rotations of objects, in Proceedings of the 4th Samara’s University Conference, Samara, Russia, 1980, pp. 127–131 (in Russian). 34. Galdi, V., Felsen, L., and Castanon, D., Quasi-ray Gaussian beam algorithm for short-pulse two-dimensional scattering by moderately rough dielectric interfaces, IEEE Trans., Vol. AP-51, No. 2, 2003, pp. 171–183. 35. Galdi, V., Felsen, L., and Castanon, D., Quasi-ray Gaussian beam algorithm for time-harmonic two-dimensional scattering by moderately rough interfaces, IEEE Trans., Vol. AP-49, No. 9, 2001, pp. 1305–1314. 36. Hansen, T. and Johansen, P., Inversion scheme for ground penetrating radar that takes into account the planar air-soil interface, IEEE Trans., Vol. GRS-38, No. 1, 2000, pp. 496–506. 37. Chew, W. and Jin, M., Perfectly matched layers in the discretized space: An analysis and optimization, Electromagnetics, Vol. 16, 1996, pp. 325–340. 38. Galdi, V., Felsen, L., and Castanon, D., Time-domain radiation from large twodimensional apertures via narrow-wasted Gaussian beams, IEEE Trans., Vol. AP-51, No. 1, 2003, pp. 78–88. 39. Lamb, J., Low-noise, high-efﬁciency optics design for ALMA receivers, IEEE Trans., Vol. AP-51, No. 8, 2003, pp. 2035–2047. 40. Wootten, A., Ed., Science With the Atacama Large Millimeter Array, ASP, San Francisco, CA, 2001, Vol. 253. 41. Chew, W. and Weedon, W., A 3-D perfectly matched medium from modiﬁed Maxwell’s equations with stretched coordinates, Microwav. Opt. Tech., Vol. 7, No. 13, 1994, pp. 599–604. 42. Jiao, D. and Jin, J., An effective algorithm for implementing perfectly matched layers in time-domain ﬁnite-element simulation of open-region EM problems, IEEE Trans., Vol. AP-50, No. 11, pp. 1615–1623. 43. Jiao, D., Jin, J., Michielssen, E., and Riley, D., Time-domain ﬁnite-element simulation of three-dimensional scattering and radiation problems using perfectly matching layers, IEEE Trans., Vol. AP-51, No. 2, 2003, pp. 296–305. 44. Jiao, D. and Jin, J., Time-domain ﬁnite element modeling of dispersive media, IEEE Microwav. Wireless Components Lett., Vol. 11, No. 5, 2001, pp. 220–222. 45. Levshin, V., Spatial Filtering in Optical Bearing Systems, Soviet Radio, Moscow, 1971 (in Russian). 46. Berenger, J., A perfectly matched layer for the absorption of electromagnetic waves, J. Comput. Phys., Vol. 144, No. 2, 1994, pp. 185–200.

Copyright 2005 by CRC Press

1598_C12.fm Page 526 Monday, October 11, 2004 3:22 PM

526

Signal and Image Processing in Navigational Systems

47. Felsen, L. and Carin, L., Diffraction theory of frequency- and time-domain scattering by weakly aperiodic truncated thin-wire gratings, J. Opt. Soc. Amer. A, Vol. 11, No. 4, 1994, pp. 1291–1306. 48. Gedney, S., An anisotropic perfectly matched layer-absorbing medium for the truncation of FDTD lattices, IEEE Trans., Vol. AP-44, No. 12, 1996, pp. 1630–1639. 49. Sacks, Z., Kingsland, D., Lee, R., and Lee, J., A perfectly matched anisotropic absorber for use as an absorbing boundary condition, IEEE Trans., Vol. AP-43, No. 12, 1995, pp. 1460–1463. 50. Iznar, A., Pavlov, A., and Fedorov, B., Optical Electronic Devices for Cosmic Apparatus, Mashinostroenie, Moscow, 1972 (in Russian). 51. Shestov, N., Detection of Optical Signal in Noise, Soviet Radio, Moscow, 1967 (in Russian). 52. Hubral, P. and Tygel, M., Analysis of the Rayleigh pulse, Geophysics, Vol. 54, No. 5, 1989, pp. 654–658. 53. Boland, J., et al., Design of a correlator for real-time video comparisons, IEEE Trans., Vol. AES-15, No. 1, 1980, pp. 63–75.

Copyright 2005 by CRC Press

1598_book.fm Page 527 Monday, October 11, 2004 1:57 PM

13 Implementation Methods of the Generalized Approach to Space–Time Signal and Image Processing in Navigational Systems

13.1 Synthesis of Quasioptimal Space–Time Signal and Image Processing Algorithms Based on the Generalized Approach to Signal Processing As shown in Chapter 12, the optimal computer calculator constructed on the basis of the generalized approach to space–time signal and image processing in the presence of noise must determine the spatial correlation function between the moving and model images in deﬁnite conditions. Digital implementation of the generalized approach requires a high computational cost that, in some cases, limits the possibilities of practical use of this approach. The use of more simple algorithms based on the generalized approach, which allows us to decrease computational costs with predetermined accuracy, is a very real problem. Many quasioptimal algorithms based on the generalized approach were constructed in heuristic ways by different approaches to techniques of signal and image processing and the presentation of results. Such a circumstance makes comparative analysis of the algorithms more difﬁcult. Let us consider the following procedure in which we are able to synthesize such algorithms and to carry out a comparative analysis. This procedure is based on the theory of signal processing and spectral analysis. The synthesis of the quasioptimal generalized space–time signal and image processing algorithm based on the generalized approach to signal processing in noise consists of the following steps: deﬁnition of the preliminary signal and image processing form, deﬁnition of the criterial correlation function, and deﬁnition of the procedure to deﬁne an extremum of the criterial correlation function. The ﬁrst and second steps have a great inﬂuence on the characteristics of the constructed algorithms. The third step is not speciﬁc to algorithms and, for this reason, we do not consider this step in the subsequent discussion. 527 Copyright 2005 by CRC Press

1598_book.fm Page 528 Monday, October 11, 2004 1:57 PM

528

Signal and Image Processing in Navigational Systems

The ﬁrst step of synthesis is that the moving image contains excessive information. If we can remove this excessive information in the absence of noise, the probability of signal and image detection and accuracy of object coordinate deﬁnition are not decreased. Reducing by decreasing the initial information image informativeness is one possible way to make computational costs less in employing the generalized space–time signal and image processing algorithms based on the generalized approach to signal processing in the presence of noise. The possible level of decrease in image informativeness is deﬁned by the predetermined noise immunity of navigational systems. The procedure of image informativeness decrease can be considered as a linear ﬁltering. In this case, the problem of synthesis of the generalized space–time signal and image processing algorithm is reduced to the problem of the deﬁnition of the ﬁlter transfer function, which ensures the required transformation of the moving image. There are some approaches to decreasing the informative excess of space–time signals and images. The most widely used technique is to reduce image resolution. Let us consider brieﬂy this technique. The image forming with lower resolution at the k-th point includes the two-dimensional image ﬁltering at the (k − 1)-th point using the low-frequency preliminary ﬁlter of the generalized receiver linear tract with further quantization of images at a half frequency of quantization at the (k − 1)-th point. The transfer function of the low-frequency preliminary ﬁlter takes the following form:1 C (ω x , ω y ) = [cos( 0.5ω x ) ⋅ cos( 0.5ω y )]n

(13.1)

where ωx and ωy are spatial frequencies and n = 1, 2, …. Sequential decrease in image resolution lies at the root of the hierarchical signal and image processing algorithms constructed based on the generalized approach to signal processing in the presence of noise. The disadvantage of this procedure is a requirement to increase the memory block because it is necessary to store a set of object images with different resolutions. In the quantization of images, the rectangular raster is often used. For this, the values of each coordinate are chosen at equidistance points with Kotelnikov’s frequency. Digitization based on the rectangular raster is a particular case of the general quantization procedure, in accordance with which the nonrectangular raster is used to choose the values of the continuous object image. In some cases, it is worthwhile to use a hexagonal raster that allows us to decrease the required number of readings to 13. 4% in comparison with the rectangular digitization. There are several complex procedures of image preprocessing that allow us to keep the geometrical images of deﬁnite forms such as the line, the circle, etc., in the preliminary ﬁltered object image. These procedures are used in the image processing of contour lines. The second step in the synthesis of the quasioptimal generalized space–time signal and image processing algorithm is the choice of the criterial function type or characteristic allowing us to estimate the degree or Copyright 2005 by CRC Press

1598_book.fm Page 529 Monday, October 11, 2004 1:57 PM

Implementation Methods of the Generalized Approach to Space-Time

529

measure based on which we can make a decision about the similarity between the compared object images [see Equation (12.68) and Equation (12.69)]. A set of probability relationship characteristics is known. This set should satisfy the following requirements. The functional measure RXY exists for some pairs of the random variables X and Y always and the probability that the random variables X and Y are not constant is equal to unity. RXY is symmetrical, i.e., RXY = RYX. The functional measure is within the limits of the interval [0,1], i.e., 0 ≤ RXY ≤ 1; if the random variables X and Y are independent the equality RXY = 0 is true. The equality RXY = 1 corresponds to the functional dependence between the random variables X and Y, i.e., X = f (Y) or Y = g(X), where f (Y) and g(X) are the measured Boreal functions. If the measured Boreal functions f (Y) and g(X) coincide with the real coordinate system axes, the following equality R [ f ( Y ), g ( X )] = rXY

(13.2)

is true. If the random variables X and Y obey the Gaussian probability distribution density, the equality RXY = |rXY| is true, where rXY is the coefﬁcient of correlation between the random variables X and Y. It should be noted that at the present time, there are no universally adopted and rigorously proven mathematical requirements for the criterial correlation functions. Therefore, we consider brieﬂy the main classes of the criterial correlation functions.

13.1.1 Criterial Correlation Functions Criterial correlation functions are based on the determination of the mutual correlation function of a stochastic process. The mutual correlation function R1.1 (α x ) = M {[a ∗ ( x ) − M [a ∗ ( x )]] ⋅ [X ( x + α x ) − M [X ( x + α x )]]} (13.3) is the mixed central moment of the second order, where a*(x) is the model image, X(x) is the moving image, and M denotes the mean. The normalized mutual correlation function can be determined in the following form: R1.2 (α x ) =

R1.1 (α x ) ∗

σ[a ( x )] ⋅ σ[X ( x )]

,

(13.4)

where σ is the operator of the mean square deviation. The functions R1.1(αx) and R1.2(αx) are functionally related in the following form: R1.1(αx) = R1.2(αx) · σ[a*(x)] · σ[X(x)].

Copyright 2005 by CRC Press

(13.5)

1598_book.fm Page 530 Monday, October 11, 2004 1:57 PM

530

Signal and Image Processing in Navigational Systems

The correlation function of the mixed initial moment of the second order is determined by R1.3 (α x ) = M{a* ( x) ⋅ X( x + α x )} .

(13.6)

The functions R1.1(αx) and R1.3(αx) are related by R 1.1(α x ) = R 1.3 (α x ) − M[a * ( x)] ⋅ M[X ( x)] .

(13.7)

The weighted criterial correlation function of the second order moment is denoted by R1.4(αx). In spite of the similar structure of the criterial functions R1.1(αx) and R1.3(αx), there are big differences in their peculiarities. Let us consider these peculiarities using the example of binary (0 and 1) image processing. In the criterial correlation function R1.3(αx), only elements of the moving image X(x) having the brightness characteristics corresponding to 1 and coinciding with elements of the model image a*(x), which also have the brightness characteristics corresponding to 1, can inﬂuence the computer-calculated results. Other elements of the moving image X(x) are multiplied by the corresponding elements of the model image a*(x) with the brightness characteristics corresponding to 0. Thus, these elements do not contribute to the criterial correlation function magnitude. The great advantage of the criterial correlation function form is that the moving images observed under various conditions of lightness give the same maximal magnitudes of the criterial correlation function R1.3(αx) under the condition that the contour lines of the moving and model images are the same. At that time, the use of the criterial correlation function R1.3(αx) is limited in navigational systems that are critical with respect to the probability of a false alarm. This is so because, even in the absence of the investigated object image, the sufﬁcient number of elements of the moving image X(x) with the brightness characteristics corresponding to 1 caused by noise and underlying background noise sources can coincide with the elements of the model image a*(x), the brightness characteristics of which also correspond to 1. Therefore, the probability of a false alarm increases. The use of R1.1(αx) allows us to decrease the probability of a false alarm, but the probability of object image detection also decreases.2 The use of the weighted criterial correlation function of the second-order moment ensures a predetermined ratio between the probability of object image detection and the probability of a false alarm R 1.4 (α x ) = R 1.3 (α x ) − βM [a * ( x)] ⋅ M[X ( x)] , where 0 ≤ β ≤ 1.

Copyright 2005 by CRC Press

(13.8)

1598_book.fm Page 531 Monday, October 11, 2004 1:57 PM

Implementation Methods of the Generalized Approach to Space-Time

531

13.1.2 Difference Criterial Functions In the general case, the difference criterial correlation function can be represented in the following form: R2.0 (α x ) = M [|a ∗ ( x ) − X ( x + α x )|p ] ,

(13.9)

where p = 1, 2,… . There are also the following difference criterial correlation functions. The average square difference criterial correlation function can be determined in the following form: R 2.1(α x ) = M [a * ( x) − X ( x + α x )]2 .

(13.10)

Reference to Equation (13.10) shows that, after squaring the ﬁrst and third terms that are the variances, the second term is the double criterial correlation function R1.3(αx). Thus, we can write R 2.1(α x ) = −2 R 1.3 (α x ) + D [a * ( x)] + D [X ( x)] .

(13.11)

The average absolute difference criterial correlation function takes the following form: R 2.2 (α x ) = M [|a * ( x) − X ( x + α x )|] .

(13.12)

Note that for binary image processing, the difference criterial correlation functions R2.1(αx) and R2.2(αx) are the same, i.e., R2.1(αx) = R2.2(αx). R2.2(αx) is connected with the normalized mutual correlation function R1.2(αx) by the relationship R 1.2 (α x ) = 1 −

R 22.2 (α x )

,

2µ 2σ[a * ( x)] ⋅ σ[X ( x)]

(13.13)

where µ is the coefﬁcient depending on the probability distribution density of image brightness characteristics. In the case of the Gaussian probability distribution density of image brightness characteristics, we can assume that µ = π2 . The peculiarity of R2.2(αx) is that there is no need to carry out the operation of centering, and obtained estimations are invariant with respect to some form of nonstationary state. The Minkovsky function3 takes the following form:

[

R 2.3 (α x ) = M | a * ( x) − X ( x + α x )|

Copyright 2005 by CRC Press

]

2

.

(13.14)

1598_book.fm Page 532 Monday, October 11, 2004 1:57 PM

532

Signal and Image Processing in Navigational Systems

The Camber function has the following form:  a∗ ( x) – X( x + α x )    R2.4 (α x ) = M  ∗ .  a ( x) + X( x + α x ) 

(13.15)

The Chebyshev function takes the following form: R2.5 (α x ) = max|a∗ ( x) − X( x + α x )|. x

(13.16)

The remarkable superiority of the difference criterial correlation functions compared to the other correlation functions is the absence of the product. Thus in using the difference criterial correlation functions, the computational cost is 4 to 10 times less compared with the use of the other correlation functions. At the same time, in using the difference criterial correlation functions, the detection performance is worse in comparison with when we use the correlation functions at low values of the signal-to-noise ratio. The difference criterial correlation functions are equal to zero in the case of exact matching between the same object images and are high in value in the case of mismatching between object images. When we use the difference criterial correlation function R2.0(αx), with an increase in the value of the parameter p, the detection performances are improved, but if the condition p ≥ 3 is satisﬁed, this improvement is negligible. Note that the difference can be replaced by summing for the criterial correlation functions R2.0(αx), …, R2.5(αx). 13.1.3 Spectral Criterial Functions The criterial correlation function in the spectral range takes the following form: R3.1 (α x ) = −1 {Sa* (ω x ) ⋅ SX (ω x )} = −1 {Sa*X (ω x )},

(13.17)

where Sa*(ωx) is the power spectral density of the model image, SX(ωx) is the power spectral density of the moving image, Sa*X(ωx) is the mutual power spectral density, –1 is the inverse Fourier transform, and the sign (·) denotes the complex conjugate value. The Rot function4 takes the following form:  S (ω )  R3.2 (α x ) = −1  a*X x  .  SX (ω x )  The function of coherency is determined by4

Copyright 2005 by CRC Press

(13.18)

1598_book.fm Page 533 Monday, October 11, 2004 1:57 PM

Implementation Methods of the Generalized Approach to Space-Time   Sa2*X (ω x ) −1 R32.3 (α x ) = −1   = {Sa*X (ω x ) ⋅ γ (ω x )} . S ( ω ) ⋅ S ( ω )  a* x X x 

533

(13.19)

The Knapp function4 takes the following form:   Sa*X (ω x )⋅|γ (ω x )|2 R3.4 (α x ) = −1  . 2 |Sa*X (ω x )|[1−|γ (ω x )|] 

(13.20)

The spectral criterial functions are based on the relationship between the mutual correlation function and the mutual power spectral density using the Fourier transform [see Equation (12.56) and Equation (12.57)]. These functions allow us to decrease the computational cost due to the use of the fast Fourier transform algorithms. Moreover, signal processing in the spectral range allows us to amplify the components of those spatial frequencies at which the signal-to-noise ratio is maximal and to compensate for components distorted by noise.

13.1.4 Bipartite Criterial Functions The bipartite criterial correlation functions suppose digital image processing if the number of quantization is equal to two or more. If each element of the ﬁrst object image with the relative object image shift αx has the number of quantization equal to i, and each element of the second object image has the number of quantization equal to j, then the bipartite criterial correlation function Fij(αx), 0 ≤ i, j ≤ 2n – 1, is increased by unity. Here 2n is the number of quantization. Consequently, under the condition i = j, the bipartite criterial correlation function Fij(αx) is equal to the number of elements having the same brightness characteristics. Under the condition i ≠ j, the bipartite criterial correlation function Fij(αx) is equal to the number of elements having different brightness characteristics. If the object images with dimensions N × N are identical, we can write 2 n −1 2 n −1

∑∑ i=0 j=0

 N 2 Fij (α x ) =   0

at

i = j,

at

i ≠ j.

(13.21)

As an example, consider two criterial correlation functions obtained by the use of bipartite criterial correlation functions in the course of multilevel quantization:5

Copyright 2005 by CRC Press

1598_book.fm Page 534 Monday, October 11, 2004 1:57 PM

534

Signal and Image Processing in Navigational Systems

1 R4.1 (α x ) = N

n−1

∑ F (α ) , ii

  2 −1   Fii α x R4.2 (α x ) =  2n − 1 i=0  Fij α x   j = 0 n

∏

(13.22)

x

i=0

( )

∑ ( )

   .   

(13.23)

The bipartite criterial correlation functions are widely used in binary image processing. For this case, let us introduce the following notation: F11 (α x ) = α ; F01 (α x ) = β; F10 (α x ) = γ ; F00 (α x ) = ζ ;

(13.24)

d* is the number of elements with brightness characteristics equal to 1 in the model image; d is the number of elements with brightness characteristics equal to 1 in the moving image; f* is the number of elements with brightness characteristics equal to 0 in the model image; f is the number of elements with brightness characteristics equal to 0 in the moving image. The wellknown bipartite criterial correlation functions3 fall into the following types: the Rao function R4.3 = 2(α + β + γ + ζ ) −1 ;

(13.25)

R4.4 = α (α + β + γ ) −1 ;

(13.26)

R4.5 = 2 α(2 α + β + γ )−1 ;

(13.27)

the Jakard function

the Deak function

the Soucal and Snit function R4.6 = α(α + 2 β + 2 γ )−1 ;

(13.28)

R4.7 = α (β + γ ) −1 ;

(13.29)

the Kulzinsky function

Copyright 2005 by CRC Press

1598_book.fm Page 535 Monday, October 11, 2004 1:57 PM

Implementation Methods of the Generalized Approach to Space-Time

535

the ﬁrst Rodgers and Tanimoto function R4.8 = α ( d ∗ + d − α ) −1 ;

(13.30)

the second Rodgers and Tanimoto function R4.9 = (α + ζ) ⋅ (α + 2 β + 2 γ + ζ)−1 ;

(13.31)

the Soucal and Mishner function R4.10 = (α + ζ ) ⋅ (α + β + γ + ζ ) −1 ;

(13.32)

R4.11 = (2ζ – βγ) · (2ζ + βγ)–1;

(13.33)

the Yule function

the Chamman function R4.12 = (α + ζ − β − γ ) ⋅ (α + β + γ + ζ ) −1 .

(13.34)

The choice of the criterial correlation functions is deﬁned by the relative importance of elements with brightness characteristics equal to 1 or 0 and by the relative importance of events that are coincident or not coincident with the brightness characteristics of elements for the considered problem. For example, the bipartite criterial correlation function R4.5 has a double weight in the case of coincidence of elements with the brightness characteristics equal to 1. The bipartite criterial correlation function R4.6 has a double weight in the case of noncoincident elements, i.e., the brightness characteristics of elements are equal to 1 and 0. If we have equivalent elements with the brightness characteristics equal to 1 and 0, it is worthwhile to use the bipartite criterial correlation function R4.3. The sign functions can be also considered as the bipartite criterial correlation functions. The normalized mutual criterial correlation function R1.2 (αx) for the stationary ergodic images obeying the Gaussian probability distribution density of brightness can be deﬁned using the probabilities of coinciding signs in the following form:6 R1.2 (α x ) = − cos 2π ⋅ Pa(∗++X ) (α x ) = − cos 2π ⋅ Pa(∗−X− ) (α x ),

(13.35)

where P (∗++ ) (α x ) is the probability of coincidence of positive signs only, and a X Pa(∗−X− ) (α x ) is the probability of coincidence of negative signs only.

Copyright 2005 by CRC Press

1598_book.fm Page 536 Monday, October 11, 2004 1:57 PM

536

Signal and Image Processing in Navigational Systems

13.1.5 Rank Criterial Functions The rank criterial correlation functions deﬁne the ranking of object image elements by the brightness characteristics. In other words, the rank criterial correlation function deﬁnes the position of the object image elements according to the order of the decreasing or increasing brightness characteristics. Each object image element is denoted by the corresponding number — the rank. The Spearman function6 is most widely used: N

∑ (r

6 R5.1 (α xj ) = 1 −

a*i + α xj

− rXi )

i=1

N( N 2 − 1)

,

(13.36)

where ra*i is the rank of the i-th element of the model image and rXi is the rank of the i-th element of the moving image. At the present time, a formalized method of choosing the type of preliminary image processing and the criterial correlation function is absent. Therefore, the synthesis of quasioptimal generalized signal and image processing algorithms based on the generalized approach to signal processing in the presence of noise is carried out only in a heuristic way.

13.2 The Quasioptimal Generalized Image Processing Algorithm Let us consider the example of the use of the procedure discussed in the previous section. In the synthesis of the quasioptimal generalized image processing algorithm, we are guided by the following main statements. First, the algorithm should operate both with object image identiﬁcation (the condition of searching) and identiﬁed object image tracking (the condition of nonsearching). In object image identiﬁcation, the main characteristics of the algorithm are the probabilities of object omission and false alarm. The accuracy of identiﬁcation is not so important and plays a secondary role. In identiﬁed object image tracking (nonsearching), ﬁrst, the quasioptimal generalized image processing algorithm should guarantee high accuracy of object image tracking because an incorrect mismatching of object images is improbable. Therefore, in object image identiﬁcation, the space–time criterial correlation function between the moving and model images must ensure a maximal ratio between the main and side lobes. In other words, this function should tend to approach the delta function in the limiting case. In object image identiﬁcation, the determination of the moving object image coordinates is carried out using a derivative of the criterial correlation function. Therefore, the main lobe must be extended a little. Second, the algorithm

Copyright 2005 by CRC Press

1598_book.fm Page 537 Monday, October 11, 2004 1:57 PM

Implementation Methods of the Generalized Approach to Space-Time

537

should ensure image processing in the spectral range, which allows us to reduce computational costs due to the use of the fast Fourier transform. Third, the algorithm has to be functionally related to the universally adopted generalized approach to signal processing in the presence of noise. In accordance with the preceding statements, the sequence of the quasioptimal generalized image processing algorithm based on the generalized approach to signal processing in the presence of noise7–10 can be represented in the following form. During the ﬁrst step, the preprocessing of the moving and model images is carried out. Its main purpose is to decrease the object image informativeness. Hereinafter, it is necessary to carry out additional processing of the moving image to obtain the criterial correlation function with the minimal width of the main lobe. We can present these two steps in the form of linear ﬁltering (see Figure 13.1). Let us introduce the following notation: h1(x, y) is the weight function and C1(ωx, ωy) is the transfer function of the ﬁlter providing a reduction in the object image informativeness; â*(x, y) and Sˆ a* (ωx , ωy) are the model image and the power spectral density of the model image, respectively, with decreased informativeness; Xˆ ( x , y ) and Sˆ X (ωx , ωy) are the moving image and the power spectral density of the moving image, respectively, with decreased informativeness; h2(x, y) is the weight function and C2(ωx , ωy ) is the transfer function, respectively, of the ﬁlter during the second step of image preprocessing (additional processing of the moving image). η(x, y)

^ y) η(x,

h 2(x, y)

h 1(x, y )

η(x, y)

AF W (x, y )

X(x, y)

^ X (x, y)

h 2(x, y)

h 1(x, y )

X(x, y) PF

× ^ a*

×

−

×

+

+ × FIGURE 13.1 The generalized receiver.

Copyright 2005 by CRC Press

a*

h 1(x, y)

−

+ +

+ Model Image

+

Integrator

− Z(x, y )

1598_book.fm Page 538 Monday, October 11, 2004 1:57 PM

538

Signal and Image Processing in Navigational Systems

Taking into consideration this notation, the algorithm takes the following form: Z( x, y) ≅

∫ ∫ {2 Xˆ (x, y) ∗ h (x, y) ⋅ aˆ (x − λ , y − λ ) *

2

x

y

X Y

− Xˆ ( x, y) ∗ h2 ( x, y) ⋅ Xˆ * ( x, y) ∗ h2 ( x, y)

,

+ ηˆ 2 ( x, y) ∗ h2 ( x, y) ⋅ ηˆ *2 ( x, y) ∗ h2 ( x, y)} dx dy = δ ( x − λ x , y − λ y ) (13.37) where aˆ * ( x − λ x , y − λ y ) = a ( x − λ x , y − λ y ) ∗ h1 ( x, y) ;

(13.38)

Xˆ ( x , y ) = X ( x , y ) ∗ h 1 ( x , y ) ;

(13.39)

ηˆ ( x , y ) = η( x , y ) ∗ h 1 ( x , y ) ;

(13.40)

the symbol (∗) denotes convolution. The formula in Equation (13.37) is true and conditional in the statistical sense. We can obtain the output statistic in the form of the delta function in processing object images unlimited in informativeness only. In practice, in real image processing, the main lobe of the criterial correlation function has a ﬁnite width and the side lobes are present. However, we can assume that the width of the main lobe and the level of the side lobes tend to approach zero. Therefore, we can assume that the algorithm can be considered as ideal, even if quasioptimal. Using the feature of convolution associability, the moving image at the output of the ﬁlter with the weight function h2(x, y) can be determined in the following form: [X(x, y) ∗ h1(x, y)] ∗ h2(x, y) = [h2(x, y) ∗ h1(x, y)] ∗ X(x, y) = hΣ(x, y) ∗ X(x, y), (13.41) where hΣ(x, y) = h2(x, y) ∗ h1(x, y) is the summing weight function of the ﬁlter. Thus, the problem of synthesis of the quasioptimal generalized image processing algorithm based on the generalized approach to signal processing in the presence of noise can be thought of as a solved problem if we are able to deﬁne the weight functions h1(x, y) and hΣ(x, y) or h2(x, y) of ﬁlters and corresponding transfer functions.

Copyright 2005 by CRC Press

1598_book.fm Page 539 Monday, October 11, 2004 1:57 PM

Implementation Methods of the Generalized Approach to Space-Time

539

Let us deﬁne the weight function h1(x, y) of the ﬁlter forming the object image with decreased informativeness. In navigational systems, it is worthwhile to have information at the center of the image scene, where the object is located, in more detail, and information at the periphery of the image scene in less detail with the purpose of reducing the effect of background noise. It is possible to solve this problem in the course of image processing using the weighted functions having the maximal value equal to unity at the center of the image scene and decreasing to zero at the periphery of image scene.11 However, the introduction of the weight functions leads to an increase in the computational costs for image processing. The ﬁlter, the output signal of which can be presented by the totality of readings reckoned on radial lines emanating from the coordinate system origin and uniformly distributed within the limits of the interval [0, 2π], can be used as an alternative method of reducing image informativeness (see Figure 13.2). Thus, the requirement to reduce the object image informativeness with predetermined nonuniformity is carried out without increase in computational costs. The transfer function of the ﬁlter ensuring image processing by the use of ﬁltering features of the delta function can be determined in the following form: ωy

ο

ο ∆

ο

ο

ο

ο ∆ ∆ο

ο

ο

∆ο ∆

∆ο ∆

∆ ο

∆ ∆ο ∆

ο

ο

∆ ∆

∆

∆ ο

∆

∆

ο

∆ ο

ο∆

ο

ο∆

∆ ο ∆ ∆ ∆

∆∆∆ ο ∆ ο ∆

∆ο ο ∆

∆

∆

ο

ο

ο

ο

∆ο

ο

∆

∆

ο ∆

ο

∆ ο ∆ ∆

ο ∆ ∆ο ∆

ο ∆

ο

∆ο

ο

∆

∆

ο

ο

ο

ο

ωx

∆

FIGURE 13.2 Radial lines: o — readings in the Cartesian coordinate system; ∆ — readings on radial lines.

Copyright 2005 by CRC Press

1598_book.fm Page 540 Monday, October 11, 2004 1:57 PM

540

Signal and Image Processing in Navigational Systems

p

n

C1 (ω x , ω y ) =

∑ ∑ δ{q (ω , ω )} ⋅ δ{u (ω , ω )}, k

x

y

m

x

(13.42)

y

k =1 m=1

where qk(ωx, ωy) is a function of the line bundle and um(ωx, ωy) is a function of concentric circles with the center at the coordinate system origin. Therefore, we assume that the well-known features of the delta function are true for the two-dimensional case: δ2(x) = cδ(x), where c is an arbitrary constant and δ(x – a)(x – b) = 0 if the condition a ≠ b is satisﬁed.12 Thus, the transfer function of the ﬁlter has the form (see Figure 13.3) n

C1 (ω x , ω y ) =

p

∑ ∑ δ {ω k =1 m=1

y

( )} {

}

− ω x ⋅ tg θ n k ⋅ δ ω − mω 0 ,

(13.43)

ωy

δ(ωx, ωy)

ωx

FIGURE 13.3 The transfer function.

Copyright 2005 by CRC Press

1598_book.fm Page 541 Monday, October 11, 2004 1:57 PM

Implementation Methods of the Generalized Approach to Space-Time

where n is the number of lines in the bundle, θ n = Kotelnikov’s frequency, p =

π n

541

; ω = ω 2x + ω 2y ; ω 0 is

; k ≠ 0.5n, and Ω is the highest frequency of the

Ω ω0

power spectral density Sa(ωx , ωy) of the signal. The power spectral density of the signal at the output of the ﬁlter with the weight transfer function C1(ωx , ωy) takes the following form: n

p

∑ ∑ δ{ω

Sˆ a (ω x , ω y ) = Sa (ω x , ω y )

y

− ω x ⋅ tg (θn k)} ⋅ δ{ω − mω 0 }

k =1 m=1

(13.44) and tends to approach the power spectral density of the initial signal as n → ∞. Now, let us consider the synthesis of the ﬁlter with the transfer function C2(ωx , ωy). Let us consider the ﬁlter functioning for the searching condition. In this case, the ﬁlter would ensure the processing of the moving image, under which the criterial correlation function is transformed into the delta function for the limiting case. The moving image at the output of the ﬁlter with the weight function h2(x, y) can be written in the following form: X ( x , y ) = Xˆ ( x , y ) ∗ h 2 ( x , y ) .

(13.45)

In the spectral range, Equation (13.45) can be written in the following form: S X (ω x , ω y ) = Sˆ X (ω x , ω y ) ⋅ C 2 (ω x , ω y ) .

(13.46)

The criterial correlation function of statistic at the generalized detector output can be written in the following form: RZ ( x , y) =

∫ ∫ {2X(x, y) ⋅ aˆ (x, y) − X(x, y) ⋅ X (x, y) + η(x, y) ⋅ η (x, y)} dx dy, ∗

∗

∗

X Y

(13.47) where η ( x , y ) = ηˆ ( x , y ) ∗ h 2 ( x , y ). In the spectral range, Equation (13.47) can be written in the following form:

Copyright 2005 by CRC Press

1598_book.fm Page 542 Monday, October 11, 2004 1:57 PM

542

Signal and Image Processing in Navigational Systems

Sz (ω x , ω y ) =

∫ ∫ {2S (ω , ω ) ⋅ Sˆ (ω , ω ) − S (ω , ω ) ⋅ S (ω , ω ) X

x

y

a*

x

y

X

x

* x

y

x

y

Ω X ΩY

}

+ Sη (ω x , ω y ) ⋅ Sη* (ω x , ω y ) dxdy , (13.48) where Sη (ω x , ω y ) = Sˆ η (ω x , ω y ) ⋅ C2 (ω x , ω y )

(13.49)

= [Sη (ω x , ω y ) ⋅ C1 (ω x , ω y )] ⋅ C2 (ω x , ω y ) ,

Sη(ωx , ωy) is the power spectral density of the reference sample noise η(x, y) at the additional ﬁlter output of the generalized detector. The noise ξ(x, y) at the preliminary ﬁlter output of the generalized detector included in the moving image X(x, y) can be presented by analogous formulae. Because we require that the criterial correlation function of statistic at the generalized detector output would tend to approach the delta function δ(x, y), then, representing the power spectral densities of the moving and model images in the complex form and substituting Equation (13.46) in Equation (13.48), we can write j (ϕ δ( x , y) = −1 {{2|Sˆ X (ω x , ω y )|e j (ϕ −|Sˆ X (ω x , ω y )|e

+|Sˆ η (ω x , ω y )|e

Xx

+ ϕXy )

⋅|Sˆ a∗* (ω x , ω y )|e

+ ϕXy )

− j ( ϕ xx + ϕ x y ) ⋅|Sˆ X∗ (ω x , ω y )|e

j (ϕ Xx + ϕ X y )

− j (ϕ Xx + ϕ X y )

Xx

⋅|Sˆ η∗ (ω x , ω y )|e

− j (ϕ * + ϕ * ) a a x

y

} ⋅ C2 (ω x , ω y )} , (13.50)

where –1(x) is the inverse Fourier transform. Based on Equation (13.50), it is not difﬁcult to ensure that the transfer function C2(ωx, ωy) should have the following form: C2 (ω x , ω y ) = 2Sˆ X (ω x , ω y ) ⋅ Sˆ a** (ω x , ω y ) − Sˆ X (ω x , ω y ) −1 × Sˆ X* (ω x , ω y ) + Sˆ η (ω x , ω y ) ⋅ Sˆ η* (ω x , ω y ) .

(13.51)

Substituting Equation (13.51) in Equation (13.50), we obtain

−1{e Copyright 2005 by CRC Press

j ( ϕ Xx − ϕ ∗ + ϕ X y − ϕ ∗ ) a a x

y

} = δ(x, y) .

(13.52)

1598_book.fm Page 543 Monday, October 11, 2004 1:57 PM

Implementation Methods of the Generalized Approach to Space-Time

543

The result obtained can be interpreted as follows. The device with the transfer function C2(ωx , ωy) is the phase ﬁlter, i.e., the ﬁlter at the output of which the phase component of the complex power spectral density of the object image does not vary and the power spectral density of the amplitude component of the object image is normalized. The fact that the phase component plays the main role in solving the pattern recognition problem is discussed in References 11 and 13.11,13 The phase component is important because the total information regarding the shift of the object image is included in it. For instance, let us consider that the object image a(x, y) with the power spectral density Sa(ωx, ωy ) of a(x, y) shifted by the value λx along the axis x, and by the value λy along the axis y, can be written in the following form: ∞ ∞

{a( x − λ x , y − λ y )} =

∫ ∫ a(x − λ , y − λ ) ⋅ e x

− j(ω x x + ω y y )

y

dx dy , (13.53)

−∞ −∞

where F denotes the Fourier transform. Let us introduce the notation u = x – λx and υ = y – λy . Then, Equation (13.53) can be written in the following form: ∞ ∞

{ a( x − λ x , y − λ y )} =

∫ ∫ a(u, υ) ⋅ e

− j [ ω x ( u + λ x ) + ω y ( υ + λ y )]

du dυ

−∞ −∞

= S(ω x , ω y ) ⋅ e

− j (ω xλ x + ω yλ y )

(13.54)

.

Reference to Equation (13.54) shows that the phase component of the complex power spectral density of the object image varies. The possibility of using only the phase component of the object image complex power spectral density is proved by the experimental investigations discussed in References 11 and 13.11,13 In the use of the nonsearching condition, the ﬁlter transfer function can be generalized by introducing the power degree in the exponent of Equation (13.51): C2 (ω x , ω y ) =|2Sˆ X (ω x , ω y ) ⋅ Sˆ a∗∗ (ω x , ω y ) − Sˆ X (ω x , ω y ) × Sˆ X∗ (ω x , ω y ) + Sˆ η (ω x , ω y ) ⋅ Sˆ η∗ (ω x , ω y )|− .

(13.55)

If = 1, the criterial correlation function takes the form of the delta function and the quasioptimal generalized image processing algorithm based on the generalized approach to signal processing in the presence of noise ensures the searching condition. In the condition 0 ≤ < 1, the delta function becomes Copyright 2005 by CRC Press

1598_book.fm Page 544 Monday, October 11, 2004 1:57 PM

544

Signal and Image Processing in Navigational Systems

fuzzy so that it ensures the nonsearching condition for the functioning of the tracking device in the navigational system, the discrimination characteristic of which is deﬁned by the derivative with decreased informativeness. In the nonsearching condition, the value of is deﬁned by the statistical characteristics of the object image and background noise.14 Thus, the quasioptimal generalized image processing algorithm can be written in the following form:

{

−1

Z(λ x , λ y ) = max  2SX (ω x , ω y ) ⋅ Sa** (ω x , ω y ) − SX (ω x , ω y )

}

× SX* (ω x , ω y ) + Sη (ω x , ω y ) ⋅ Sη* (ω x , ω y )  ,

(13.56)

where S X (ω x , ω y ) = S X (ω x , ω y ) ⋅ C Σ (ω x , ω y ) ;

(13.57)

Sˆ a∗∗ (ω x , ω y ) = S a∗∗ (ω x , ω y ) ⋅ C 1 (ω x , ω y ) ;

(13.58)

S η (ω x , ω y ) = S η (ω x , ω y ) ⋅ C Σ (ω x , ω y ) ;

(13.59)

n

C1 (ω x , ω y ) =

p

∑ ∑ δ {w

y

} {

}

− ω x ⋅ tg(θn k) ⋅ δ ω – mω 0 ;

k =1 m=1

(13.60)

CΣ (ω x , ω y ) = n

p

∑ ∑ δ{ω

y

− ω x ⋅ tg (θn k)} ⋅ δ{ω − mω 0 }

;

k =1 m=1

|2SˆX (ω x , ω y ) ⋅ Sˆ a∗ (ω x , ω y ) − SˆX (ω x , ω y ) ⋅ SˆX∗ (ω x , ω y ) + Sˆ η (ω x , ω y ) ⋅ Sˆ η∗ (ω x , ω y )|− ∗

(13.61) 0 ≤ ≤ 1; k ≠ 0.5n; p =

Ω π ; and θn = . ω0 n

(13.62)

The formula in Equation (13.56) deﬁnes the quasioptimal generalized image processing algorithm for the searching condition. In the nonsearching condition, the difference is that the shift parameters λx and λy are deﬁned by the zero value of the ﬁrst-order partial derivatives with respect to λx and λy and by the total derivatives with respect to time. As n → ∞ and at = 0, the algorithm corresponds to the universally adopted generalized signal Copyright 2005 by CRC Press

1598_book.fm Page 545 Monday, October 11, 2004 1:57 PM

Implementation Methods of the Generalized Approach to Space-Time

545

processing algorithm based on the generalized approach to signal processing in the presence of noise. This fact veriﬁes the quasioptimality of the generalized image processing algorithm given by Equation (13.56). Let us consider the practical realization of the quasioptimal generalized image processing algorithm based on the generalized approach to signal processing in the presence of noise. In practice, it is very difﬁcult to realize the algorithm given by Equation (13.56). The difﬁculties are caused by the fact that in object image processing by a ﬁlter with the transfer function C1(ωx , ωy) given in the spectral range by concentric circles, we assume that the power spectral density of the object image is deﬁned in the polar coordinate system u and θ, not in the Cartesian coordinate system ωx and ωy . In this case, the power spectral density of the object image can be determined in the following form:14 ∞

∑C

(u) ⋅ e − jnθ

(13.63)

∫ ρ C (ρ)J (pu)dρ ,

(13.64)

Sa (u, θ) = 2 π

nn

n= − ∞

where ∞

Cnn (u) =

n

n

−∞

1 Cn (ρ) = 2π

π

∫ a(ρ, φ) ⋅ e

− jnφ

dφ .

(13.65)

−π

Equation (13.63) represents the power spectral density of the object image a(ρ, φ) in using the Fourier-series expansion. The coefﬁcients of the Fourierseries expansion are the Chancel transform of the n-th order for the sequence Cn(ρ) = 0, where n = 0, ± 1, ± 2…. The theorem of readings has a more complex form using polar coordinates. The object image a(ρ, φ) having the power spectral density Sa(u, θ) = 0 at ρ ≥ λ can be reconstructed by its readings in accordance with Stark and Woods:15 ∞

a(ρ, φ) =

2K

∑ ∑ a(α

0i

, 22Kπ+k1 ) ⋅ θ0i (ρ) (2 K + 1)−1 + 2(2 K + 1)−1

i=1 k = 0

∞

×

2K

∑ ∑ ∑ a(α i=1 k = 0 n=1

Copyright 2005 by CRC Press

(13.66)

K

ni

, 22Kπ+k1 ) ⋅ θni (ρ) cos  n(φ −

2 πk 2K +1

)  ,

1598_book.fm Page 546 Monday, October 11, 2004 1:57 PM

546

Signal and Image Processing in Navigational Systems

where

θ ni (ρ) =

2α ni Jn (ρλ ) λJn+ 1 (α ni λ ) ⋅ (α 2ni − ρ 2 )

,

(13.67)

z

n α ni = λi , z ni is the i-th zero of the Bessel function Jn(x). Reference to Equation (13.63) and Equation (13.66) shows that various transforms of the object image a(ρ, φ) given in polar coordinates include operations with the Bessel functions and, in particular, the Chancel transform. The computational effectiveness of the Chancel transform is less than that of the fast Fourier one.16 Therefore, we consider the possibility of excluding the Chancel transform from Equation (13.67). For this purpose, we use the sequential determination of the correlation functions of rows and columns of the two-dimensional rectangular database. Thus, the transition from processing of the two-dimensional functions to one-dimensional ones guarantees a decrease in computational costs. Let us use this technique for the considered case. The totality of values of spectral components lying along the line of the bundle q(ωx , ωy) and given by Equation (13.45) and Equation (13.46) is chosen as a partial one-dimensional deﬁnition of the object image. Note that these spectral components are the central cross section of the power spectral density of the object image, which are uniformly distributed within the limits of the interval θ ∈ [0, 2π]

a (ω u , θ) = Sa (ω x , ω y ) ω x = ω u cos θ . ω y = ω u sin θ

(13.68)

The power spectral density of the object image is shown in Figure 13.4. Its cross sections at θ = 0° and at θ = 90° are shown in Figure 13.5 and Figure 13.6, respectively. Obviously, with an increase in the number n of central cross sections of the power spectral density, we can deﬁne the total power spectral density of the object image with predetermined accuracy and, consequently, we can deﬁne the space–time correlation function of the moving and model images. Therefore, Equation (13.56) deﬁning the synthesized quasioptimal generalized image processing algorithm based on the generalized approach to signal processing in the presence of noise is divided into n analogous formulae, each of which is derived from the correlation function determined under the same angles of the central cross sections of the moving and model image power spectral densities. The zero value of the derivative of the space–time correlation function deﬁnes the shift αθ of cross sections and with it the shift of compared images along the direction θ. To deﬁne values λx and λy of the object image shift in the Cartesian coordinate system, it is necessary to deﬁne the correlation function of two cross sections of the power spectral densities

Copyright 2005 by CRC Press

1598_book.fm Page 547 Monday, October 11, 2004 1:57 PM

Implementation Methods of the Generalized Approach to Space-Time

547

ωx

ωy FIGURE 13.4 The power spectral density of the object image.

ωx

ωy FIGURE 13.5 The central cross section of the power spectral density of the object image, θ = 0°.

under arbitrary angles θk and θ. The shift is deﬁned by formulae in the condition θk < θ

Copyright 2005 by CRC Press

1598_book.fm Page 548 Monday, October 11, 2004 1:57 PM

548

Signal and Image Processing in Navigational Systems

ωx

ωy FIGURE 13.6 The central cross section of the power spectral density of the object image, θ = 90°.

 α θk + θ k + 2πn,  α = ± arccos ν   α θ   ν = cos(α − θ ) , 

(13.69)

where ν=

λy λx = . cosα sin α

(13.70)

With low values of θ and α, the approximate solution of Equation (13.69) takes the form

λ x = λ y = α θk − θ k

α θk − α θ θk − θ

.

(13.71)

Let us consider the possibility of deﬁning the central cross sections of the power spectral densities because the use of the synthesized quasioptimal generalized image processing algorithm based on the generalized approach to signal processing in the presence of noise assumes the use of central cross sections of the power spectral densities in practice. The power spectral density of the object image a(x, y) can be written in the following form: ∞ ∞

Sa (ω x , ω y ) =

∫ ∫ a(x, y) ⋅ e

−∞ −∞

Copyright 2005 by CRC Press

− j (ω x x + ω y y )

dx dy.

(13.72)

1598_book.fm Page 549 Monday, October 11, 2004 1:57 PM

Implementation Methods of the Generalized Approach to Space-Time

549

Let us assume that θ = 0°, i.e., let us consider the cross section Sa(ωx) of the power spectral density Sa(ωx, ωy), which can be determined in the following form:

a (ω x ) = Sa (ω x , ω y )

ωy = 0

.

(13.73)

Substituting Equation (13.72) in Equation (13.73), we obtain ∞ ∞

a (ω x ) =

∫ ∫ a(x, y) ⋅ e

− jω x x

dx dy .

(13.74)

−∞ −∞

Changing the order of integration in Equation (13.74), we can write ∞    − jω x x a ( x , y ) dy dx .  ⋅e   − ∞ − ∞  ∞

a (ω x ) =

∫ ∫

(13.75)

Let us denote the integral in braces in Equation (13.75) as sa(x). It is easy to see that the central cross section of the power spectral density a(ωx) is the Fourier transform ∞

s a (x ) =

∫ a (x , y ) dy.

(13.76)

−∞

The function sa(x), being the result of the transform given by Equation (13.76), is called the projection. We can rewrite Equation (13.76) in the following form: N

s a (i ) =

∑ a (i , j ),

(13.77)

j =1

where N is the number of elements in the expansion of the object image a(i, j) along the axis j. Reference to Equation (13.77) shows that the projection sa(i) is the result of summing the brightness characteristics of elements of the object image a(i, j) over the coordinate j. Thus, if the object image a(i, j) is given in the right matrix form, we can obtain the projection of this object image at the angle θ = 0° by summing the brightness characteristics of the elements of the matrix by columns. The effectiveness of the computer-calculated projection is evident. In accordance with Equation (13.75), it is necessary to determine the Fourier transform of the projection to obtain the central cross section of the power spectral density of the object image. The Copyright 2005 by CRC Press

1598_book.fm Page 550 Monday, October 11, 2004 1:57 PM

550

Signal and Image Processing in Navigational Systems

relationship between the cross sections and projections of the power spectal density of the object image is deﬁned by the theorem about projections and cross sections of multidimensional signals, the meaning of which is the following: (N − 1)-dimensional Fourier transform with respect to the projection sax1(x2 , …, xN) is the central cross section of the N-dimensional Fourier transform with respect to the function f(x1, …, xN) or, in other words, the power spectral density of the projection is the cross section of the power spectral density of the object image.17 Equations (13.72)–(13.75) are variants of this theorem about projections and cross sections in the case of twodimensional signals. A function of the cross section of the power spectral density and projections (the Fourier transform) is used in tomography and molecular biology to reconstruct multidimensional signals using their projections.17 Projections are also used to solve pattern recognition problems.4 The formula in Equation (13.76) deﬁnes the projection of the object image at θ = 0°. Let us deﬁne the projection sa(x, θ) for the arbitrary value of the angle θ within the limits of the interval [0, 2π].Note that the angle θ is a parameter, not a variable. For this purpose, we can introduce new variables or rotate the object image a(x, y) using the previous coordinate system. Rotating the object image, we can write ∞

s( x , θ) =

∫ a[(x, y)A] dy

(13.78)

−∞

where

A=

cosθ

sin θ

− sin θ

cosθ

.

(13.79)

It is well known that if a(x, y) and Sa(ωx, ωy) form a pair of Fourier transforms, then a[(x, y)A] and Sa[(ωx , ωy)A] are also a pair of Fourier transforms, if the matrix A is orthogonal, i.e., AT = A–1.18 The rotation matrix A is orthogonal. Because of this, the rotation of the cross section of the power spectral density of the object image by the angle θ corresponds to the rotation of the projection s(x, θ) by the same angle. Figure 13.7 shows us the determination of the projection s(x, θ) for an arbitrary angle θ. The elements of the projection s(k, θ) are equal to the sum of the brightness characteristics of the object image elements a(i, j) within the limits of the region dy or of the beam width b(k, θ), which is limited by lines yk and yk-1 or xk and xk-1. At θ = 0° and θ = 90°, the set of lines takes the following form: x k = k + 0.5,

Copyright 2005 by CRC Press

k = 0, 1, 2, … , N.

(13.80)

1598_book.fm Page 551 Monday, October 11, 2004 1:57 PM

Implementation Methods of the Generalized Approach to Space-Time

551

y

b(k, θ)

. . . . (1, 1) .

yk −1

. . . . .

yk

. . . . .

. . . . .

.(N, N ) . . . . θ

x

FIGURE 13.7 Projection determination.

The elements of the projection are determined by N

s( k , θ) =

∑ a(i, j),

(13.81)

j=1

where k = 1, 2, … , N; xk-1 ≤ i ≤ xk , i.e., the summation is carried out by columns of object image elements. At θ = 90° and θ = 270° the totality of lines takes the form y k = k + 0.5,

k = 1, 2, … , N.

(13.82)

The elements of the projection are determined by N

s( k , θ) =

∑ a(i, j), i=1

Copyright 2005 by CRC Press

(13.83)

1598_book.fm Page 552 Monday, October 11, 2004 1:57 PM

552

Signal and Image Processing in Navigational Systems

where k = 1, 2, … , N; yk-1 ≤ j ≤ yk , i.e., the summation is carried out by rows of object image elements. For all other angles θ, the totality of lines yk separating the beams b(k, θ) can be written in the following form according to Reference 19:19 yk = x tg (θ + 0.5π) + y0 + k|sin θ|−1 ,

k = 0, 1, 2 , … , Cθ ,

(13.84)

where y 0 = 0.5( N + 1) − L |sin θ |−1 − 0.5( N + 1) tg (θ + 0.5π);  0.5N + Int [0.5(N − 1) ⋅ (|sin θ|+ |cos θ |−1) + 0.5] , L=  0.5N + Int [0.5(N − 1) ⋅ (|sin θ|+ |cos θ |−1) + 0.5] − 1,

(13.85)

at [.] > Int [.] , at

[.] = Int [.] , (13.86)

at [.] > Int [.] ,  N + 2 Int [0.5(N − 1) ⋅ (|sin θ|+ |cos θ |−1) + 0.5] , Cθ =   N + 2 Int [0.5(N − 1) ⋅ (|sin θ|+ |cos θ |−1) + 0.5] − 1, at [.] = Int [.] , (13.87) where Int[.] is the integer part of [.]. The projection elements for an arbitrary angle θ are determined by s( k , θ) =

∑

a(i, j) ,

k = 1, 2 , … , Cθ .

(13.88)

( i , j ) ∈b ( k ,θ )

As is well known from the theory of multidimensional signals, there is a ﬁnite set of rational angles {θ1, … , θn} among inﬁnite sets of projection angles, the use of which ensures an unambiguous reconstruction of signals with minimal numbers of projections. Therefore, we deﬁne these projection angles. The rational angles can be determined by20 θ q = arctg

Tq Qq

,

(13.89)

where q = 1, … , n; T and Q are integer coprime numbers satisfying the conditions |T| ≤ N and |Q|≤ N, and n is the number of projection angles. The rational angles are within the limits of the interval [–0.5π, 0.5π]. Because the projections are periodic functions of the angle θ with the period π, i.e.,

Copyright 2005 by CRC Press

1598_book.fm Page 553 Monday, October 11, 2004 1:57 PM

Implementation Methods of the Generalized Approach to Space-Time s ( u, θ ) = s[( −1) k u, θ + kπ] ,

553

(13.90)

we can use the interval [0, π] to determine the projections. Note that the rational angles θ2q-1 and θ2q(θ = 1, … , n) are shifted with respect to each other by 0.5π. This allows us to simplify Equation (13.69) and Equation (13.71) in the deﬁnition of the estimation λx and λy in the course of determining the criterial correlation function of projections at the angles θ2q-1 and θ2q. In this case, the estimations of the object image shift along the q-th pair of orthogonal projections in the Cartesian coordinate system and take the form λ x ( q ) = α θ2q −1 cosθ 2q−1 − α θ2q sin θ 2q−1 ,

(13.91)

λ y ( q ) = α θ2q −1 sin θ 2q−1 + α θ2q cosθ 2q−1.

(13.92)

Equation (13.91) and Equation (13.92) are shown in Figure 13.8, in which an example of the processing of the object image given by the point a(x, y) = δ(x, y) is illustrated. Final estimations of the shifts λ*x and λy* of the object image are formed due to statistical averaging of the estimations λx(q) and λy(q) for all q = 1, … , n. The performed analysis of possibilities of using the quasioptimal generalized image processing algorithm based on the generalized approach to signal processing in the presence of noise shows the appropriateness of the

X (x, y) y

a*(x − λx , y − λy)

λy λθ 2q −1

s a*(k, θ2q )

λ θ 2q x s a*(k, θ 2q −1)

λx

sX (k, θ2q )

λθ 2q −1 θ2q = θ2q −1 + 0.5π

λθ2q

sX (k, θ 2q −1)

FIGURE 13.8 An example of image processing.

Copyright 2005 by CRC Press

θ2q −1

1598_book.fm Page 554 Monday, October 11, 2004 1:57 PM

554

Signal and Image Processing in Navigational Systems

sequence of steps discussed in the following text. The ﬁrst is forming an ensemble from the n projections sa(k, θq ) of the model image and sX(k, θq ) of the moving image: s( k , θq ) =

∑

(13.93)

a(i, j).

( i , j ) ∈b ( k ,θq )

The elements (i, j) ∈ b(k, θq) are given by Equation (13.84) — Equation (13.87). The projection angles θq are deﬁned in accordance with Equation (13.89). The second is the determination of the central cross sections of the object image power spectral densities N

(ω u , θq ) =

∑ s(k, θ ) ⋅ e q

− j 2 π kω u N

=| (ω u , θq )|⋅ e

jφ( ω u ,θq )

.

(13.94)

k =1

The third is the determination of the correlation functions of the moving and model image projections carried out according to R( k , θq ) = −1 {2| X (ω u , θq ) ⋅ ∗a∗ (ω u , θq )|(1− ) ⋅ e

j[ ϕ X ( ω u , θq ) − ϕ ∗ ( ω u , θq )] a

}, (13.95)

where –1 is the inverse Fourier transform, 0 ≤ < 1. The fourth is that in the searching condition, we have = 1; in the nonsearching condition, we have 0 ≤ < 1. Using the maximal value of R(k, θq) for the searching condition, the estimations αθq of the shift along the direction θq are deﬁned. Fifth, the estimations λx(q) and λy(q) of the object image shift in the Cartesian coordinate system are deﬁned by each pair of orthogonal cross sections of the power spectral densities λ x (q) = α θ2 q−1 cos θ2 q−1 − α θ2 q sin θ2 q−1

and (13.96)

λ y (q) = α θ2 q−1 sin θ2 q−1 − α θ2 q cos θ2 q−1 .

Sixth, the ﬁnal estimations of the moving image shift are determined by n

∗ x

λ (q) = n

−1

∑ λ (q) x

q=1

n

and

∗ y

λ (q) = n

−1

∑ λ (q). y

(13.97)

q=1

The quasioptimal generalized receiver is a multichannel tracking device constructed based on the generalized approach to signal processing in the presence of noise. Each channel of the receiver is processed by the central

Copyright 2005 by CRC Press

1598_book.fm Page 555 Monday, October 11, 2004 1:57 PM

Implementation Methods of the Generalized Approach to Space-Time

555

cross sections of the power spectral densities of moving and model images. The discussed quasioptimal generalized image processing algorithm allows us to reduce computational costs signiﬁcantly. This effect is caused by reducing the initial information Nn times due to formation of one-dimensional projections and use of the fast Fourier transform. The coefﬁcient of the computational cost reduction in comparison with the universally adopted generalized signal processing algorithm can be determined in the following form:

χ=

N3 . n[12 log 2 N + 0.1( N − 1)]

(13.98)

To determine the coefﬁcient χ, it is worthwhile to use the procedure discussed in Clary and Russell.21

13.3 The Classical Generalized Image Processing Algorithm In the use of the classical generalized image processing algorithm based on the generalized approach to signal processing in the presence of noise, we use the criterial correlation function. Image preprocessing is carried out by the ﬁlter with the transfer function C(ωx , ωy) = 1. In this case, we should deﬁne the correlation function maximum of the generalized receiver output statistic (see Section 12.3). The disadvantage of the use of the classical generalized image processing algorithm is the high computational cost because the determination of the correlation function of the generalized receiver output statistic is carried out for all possible relative shifts of observed object images. The total number of product operations in processing object images with dimensions M × M and N × N can be determined in the following form:21

1 = N( N + 1) ⋅ M( M + 1) .

(13.99)

This is a very large number. In the use of the Fourier transform with respect to the classical generalized image processing algorithm, we should use the criterial correlation function given by Equation (13.17). In some cases, we can obtain deﬁnite advantages, for example, if a navigational system uses optimal analog devices. In the use of the fast Fourier transform, the number of product operations can be determined by21

2 = 12 MN log 2 N + 4 M ,

Copyright 2005 by CRC Press

M ≥ N,

(13.100)

1598_book.fm Page 556 Monday, October 11, 2004 1:57 PM

556

Signal and Image Processing in Navigational Systems TABLE 13.1 Number of Product Operations Object Image Dimensions

Analog Generalized Image Processing Algorithm

Generalized Image Processing Algorithm With Fast Fourier Transform

× × × × × × ×

400 5184 73,984 1,115,136 17,305,600 272,646,144 4,328,587,264

392 2320 12,320 61,504 295,042 1,376,518 6,291,960

4 8 16 32 64 128 256

4 8 16 32 64 128 256

and we are able to decrease computational costs signiﬁcantly. The number of product operations in the use of the classical generalized image processing algorithm based on the generalized approach to signal processing in the presence of noise both for signal processing in time and in the spectral range using the fast Fourier transform is shown in Table 13.1. Table 13.1 shows that the use of the fast Fourier transform allows us to decrease the computational cost signiﬁcantly during image processing. Let us estimate the probability of true and false location of object images with reference to a control point. Let us consider the case of the matched one-dimensional object images i = 1, 2 , … , N

f (i) ,

and

g(i) ,

i = 1, 2 , … , M ,

(13.101)

where M > N and g ( i + r ) = f ( i ), i = 1, 2, … , N , the parameter r corresponds to complete matching between the model image and the moving image. Let us introduce the following notation 1 σ = N 2 f

N

∑{ i=1

f (i) −

1 N

N

∑ f (j)} ; 2

(13.102)

j=1

σn2 is the variance of the Gaussian background noise with zero mean; σ 2g ( r ) =

ρ fg ( r ) =

Copyright 2005 by CRC Press

N 1 N j=1

1 N

N

N

i=1

j=1

∑{g(i + r ) − N1 ∑ g(j + r )} ; 2

(13.103)

N

N

Σ{ f (j) − N1 iΣ= 1 f (i)} ⋅{g(j + r ) − N1 iΣ= 1 g(i + r )} σ f σ g (r )

(13.104)

1598_book.fm Page 557 Monday, October 11, 2004 1:57 PM

Implementation Methods of the Generalized Approach to Space-Time

557

is the correlation coefﬁcient in the course of matching the model image and the moving image, where r = 0, 1, … , M – N. The low boundary of the true location probability of the object image with reference to the control point in the use of the classical generalized image processing algorithm based on the generalized approach to signal processing in the presence of noise can be deﬁned using the procedure discussed in Stubberud22 N

Ptr = 1 −

∑[1 − P(z < i=1

M { Yi } σ { Yi }

)] ,

(13.105)

where z is the Gaussian random variable with zero mean and the variance equal to unity; M { Yi } = [1 − ρ fg ( r + i )](1 −

σ { Yi } =

1 N

{0.5[1 − ρ fg (r + i )]2 σσ

4 n 4 f

σ 2n σ 2f

);

(13.106)

σ2

+ [1 − ρ fg ( r + i )] σn2 } ,

(13.107)

f

i = − r , − r + 1, ... , M − N − r , i ≠ 0. The probability of false location of the object image with reference to the control point can be deﬁned based on the procedure discussed in Rockmore23 in the following form. It can be considered as the totality of the false location probabilities of the object image with reference to the control point Pfr′ (∆x, ∆y) for all possible shifts of the object image Mi2

Pfr = 1 −

∏ [1 − P′ (∆x, ∆y)] , fr

(13.108)

∆x , ∆y

where M i2 is the number of possible relative shifts. The determination of the probability of the event that the correlation function R(∆x, ∆y) formed at the generalized receiver output does not exceed the threshold Kg can be carried out using the Edgeworth-series expansion: Pr{R( ∆x , ∆y) ≤ K g } = 1 − T( x) + ϕ( x){Ni−1 [0.667 Γ 3 ( x 2 − 1)] + Ni−2 [0.041Γ 4 ( x 3 − 3 x) + 0.0014Γ 23 ( x 5 − 10 x 3 + 15 x)] + Ni3 [0.008Γ 5 ( x 4 − 6 x 2 + 3) + 0.007 Γ 3 Γ 4 ( x 6 − 15 x 4 + 45 x 2 − 15) + 0.00013Γ 33 ( x 8 − 28 x 6 + 210 x 4 − 420 x 2 + 105]} , (13.109)

Copyright 2005 by CRC Press

1598_book.fm Page 558 Monday, October 11, 2004 1:57 PM

558

Signal and Image Processing in Navigational Systems

where ∞

∞

ϕ(t)dt e ∫ 2π ∫ 1

T ( x) =

−∞

x=

Γ3 =

µ3 µ

3 2

;

Γ4 =

Ni µ2

µ

2 2

(13.110)

dt ;

−∞

K g − N i2µ 1

µ 4 − 3µ 22

− 0.5 t 2

;

(13.111)

;

and

Γ5 =

µ 5 − 10µ 2µ 3 µ 52 N i2

(13.112)

are the average number of series-expansion elements of the object image, µ1 is the average value of R(∆x, ∆y), and µn is the central moment of the n-th order of the correlation function R(∆x, ∆y). The main peculiarity of this technique is the following. First, there is no limitation on the shape of the correlation function in the case of the Gaussian probability distribution density. Second, because the false location probability of the object image with reference to a control point is determined for each relative shift of the object image, we can take into consideration variations of the statistical characteristics of the object image as a function of the shift. Let us consider the classical generalized image processing algorithm based on the generalized approach to signal processing in the presence of noise. We can represent the moving and model images in the following form: X ( x ) = Pd ( x ) + n( x )

and

a (x ) = P (x ) ,

x ∈M ,

(13.113)

where x = (x1, x2) is a two-dimensional variable. The differences between the moving image and the model image are caused by the additive noise n(x) and geometrical distortions, which can be presented by the afﬁne transform of object image coordinates24–26 Pd ( x ) = P ( Ax + t0 ),

(13.114)

where

A =α

Copyright 2005 by CRC Press

cosθ sin θ − sin θ cosθ

(13.115)

1598_book.fm Page 559 Monday, October 11, 2004 1:57 PM

Implementation Methods of the Generalized Approach to Space-Time

559

is the matrix of the relative shift of the object image by the angle θ, and α is the coefﬁcient of changes in scale. The average maximum of the correlation function of the generalized receiver output statistic, which is normalized to the average maximum of this correlation function if noise is absent, depends only on the brightness characteristics of distortions and at low values of θ and |1 – α| takes the following form: d = (1 − α ) 2 + θ 2 .

(13.116)

The normalized average maximum of the correlation function of the generalized receiver output statistic as a function of the normalized width η of the object image is shown in Figure 13.9 for various values of the distortions d. The quality of the classical generalized image processing algorithm is characterized by the ratio between the average maximum and the mean square deviation of the correlation function in the range of side lobes κ=

M { R( t0 )} σ 2 { R( t )}

.

(13.117)

The dependence κ on η is shown in Figure 13.11. Reference to Figure 13.10 shows that there is an optimal dimension of the model image for various values of distortions. The dependence of the false location probability of the R max 1.0

1

0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2

2

0.1 0.0

3 η 10 20 30 40 50 60 70 80 90 100

FIGURE 13.9 The normalized average peak of the correlation function as a function of the normalized object image width: (1) d = 0 (θ = 0°); (2) d = 0.087 (θ = 5°); (3) d = 0.174 (θ = 10°).

Copyright 2005 by CRC Press

1598_book.fm Page 560 Monday, October 11, 2004 1:57 PM

560

Signal and Image Processing in Navigational Systems 10 lgκ 1

16 14 12 10 8 6

2

4

3

2 4 0

η

10 20 30 40 50 60 70 80 90 100

FIGURE 13.10 The dependence r on the normalized object image width: (1) d = 0 (θ = 0°); (2) d = 0.087 (θ = 5°); (3) d = 0.122 (θ = 7°); (3) d = 0.174 (θ = 10°).

lgPfr 16

1

14 12 10

2

8 6 4 2 0

3 4 η 10 20 30 40 50 60 70 80 90 100

FIGURE 13.11 The probability of false location of the object image with reference to the control point as a function of the normalized object image width: (1) d = 0 (θ = 0°); (2) d = 0.087 (θ = 5°); (3) d = 0.122 (θ = 7°); (3) d = 0.174 (θ = 10°).

object image with reference to the control point on the dimensions of the object image is shown in Figure 13.11. Thus, we arrive at the following conclusions. If there are geometrical distortions of the object image, then there is an optimal dimension of the model image allowing us to minimize the false location probability of the object image with reference to the control point. This optimal dimension is proportional to the correlation function width and decreases with an increase in geometrical distortions. In the general case, the minimization of the false Copyright 2005 by CRC Press

1598_book.fm Page 561 Monday, October 11, 2004 1:57 PM

Implementation Methods of the Generalized Approach to Space-Time

561

location probability of the object image with reference to the control point must be carried out by the choice of both the dimensions and the shape and orientation of the model image. For the discussed case, we assume that the correlation function of the moving image takes a circular symmetry, because of which the optimal model image must be of square form. There are analogous dependencies in the minimization of the mean square deviation of matching between the moving and model images. However, with the given level of distortions, the dimensions of the object image in which the error of matching is minimal, is less than the dimensions of the object image that are required to minimize the false location probability of the object image with reference to the control point.

13.4 The Difference Generalized Image Processing Algorithm The difference generalized image processing algorithm based on the generalized approach to signal processing in the presence of noise uses the criterial function in the form R2.0 … R2.5 (see Section 13.12). Image preprocessing is carried out by the ﬁlter with the transfer function C(ωx , ωy) = 1, i.e., the difference generalized image processing algorithm is based on the elementby-element determination of differences of the object image brightness characteristics. Let us consider the algorithm given by Equation (13.10) in more detail. Note that using the analogous formula, we can deﬁne the distance in Euclidean space. The formula given by Equation (13.10) can be rewritten in the following form: 1 R2.1 (m, n) = 2 N

N

N

∑ ∑  a i=1 j=1

*2

(i, j) − 2 a* (i, j)X(i – m, j − n) + X 2 (i − m, j – n)  , (13.118)

where summation is carried out over all i and j, in which the arguments of the moving image are in the domain of deﬁnition. The ﬁrst term in Equation (13.118) is constant for all m and n and can be included into the threshold value in the decision making. In terms of the reference noise η(i, j) forming at the output of the additional ﬁlter of the generalized receiver, the criterial function can be determined in the following form:

– R2.1 (m, n) =

1 N2

N

N

∑ ∑ 2a (i, j)X(i − m, j − n) − X (i − m, j – n) + η (i, j) . *

2

2

i=1 j=1

(13.119)

Copyright 2005 by CRC Press

1598_book.fm Page 562 Monday, October 11, 2004 1:57 PM

562

Signal and Image Processing in Navigational Systems

The formula in Equation (13.119) is the criterial correlation function of the classical generalized image processing algorithm based on the generalized approach to signal processing in the presence of noise (see Chapter 12). Consequently, the main characteristics of the difference generalized image processing algorithm are not signiﬁcantly different from the main characteristics of the classical generalized image processing algorithm.

13.5 The Generalized Phase Image Processing Algorithm In the use of the generalized phase image processing algorithm, the criterial correlation function can be determined in the following form: 1 R( k) = N

N −1

⋅ S (m)|} ∑{{2|S (m)||

(1− L )

a*

X

⋅ e j [ 2 πmkN

−1

+ ϕ a* ( m ) − ϕ X ( m)]

m= 0

= {|SX (m)|| ⋅ SX∗ (m)|}(1−−L) − {|Sη (m)|| ⋅ Sη∗ (m)|}(1−L) × e j [ 2 πmkN 1 ≅ N

−1

+ ϕ X ( m )− ϕ a* ( m)]

N −1

⋅ S (m)|} ∑ {2|S (m)||

(1− L )

a*

(13.120)

} X

⋅ e j[ 2 πmkN

−1

+ ϕ a* ( m) − ϕ X ( m )]

,

m= 0

where

|S (m )|=

Re{S ( m )} 2 + Im{S ( m )} 2 ;

ϕ(m) = arctg

Im{S(m)} ; Re{S(m)}

(13.121)

(13.122)

N −1

Re{S(m)} =

∑ S(k) cos 2πNkm ;

(13.123)

k=0

N −1

Im{S(m)} = −

∑ S(k)sin 2πNkm ;

(13.124)

k=0

N is the number of elements of image expansion, and 0 ≤ L ≤ 1. At L = 0, we obtain the classical generalized image processing algorithm: Copyright 2005 by CRC Press

1598_book.fm Page 563 Monday, October 11, 2004 1:57 PM

Implementation Methods of the Generalized Approach to Space-Time

{

R(λ x ) = −1 2|Sa* (ω x )|| ⋅ SX∗ (ω x )|⋅e –|SX (ω x )|| ⋅ SX∗ (ω x )|⋅e

563

j ϕ a* ( ω x ) − ϕ x ( ω x ) 

j ϕ X ( ω x ) − ϕ X ( ω x ) 

+|Sη (ω x )|| ⋅ Sη∗ (ω x )|⋅e j[ϕ

X ( ω x ) − ϕ X ( ω x )]

} = −1{2Sa (ω x ) ⋅ SX∗ (ω x )}. ∗

(13.125) At L = 1, Equation (13.120) corresponds to the generalized phase image processing algorithm consisting of the following operations: the deﬁnition of the discrete Fourier transform with respect to the moving and model images, forming of the phase difference matrix at each spatial frequency by the deﬁnition of the mutual energy spectrum divided by the absolute value, the deﬁnition of the inverse Fourier transform with respect to the normalized energy spectrum, and the deﬁnition of the criterial correlation function maximum. At 0 < L < 1, Equation (13.120) corresponds to the modiﬁed generalized phase image processing algorithm. The total use of the information included in the phase component of the complex power spectral density of the signal and the partial use or modiﬁcation of the amplitude component information of the signal are characteristic of the generalized phase image processing algorithm. The phase component of the complex power spectral density of the signal is caused, in particular, by the fact that all information with respect to the relative shift of the object image is concentrated in the phase component. For example, let us consider the object image a(x, y) having the power spectral density Sa(ωx , ωy). The power spectral density of the object image a(x, y) shifted by the value λx along the x axis and by the value λy along the y axis can be written in the following form: ∞ ∞

[ a( x − λ x , y − λ y )] =

∫ ∫ a(x − λ , y − λ ) ⋅ e x

− j(ω x x + ω y y )

y

dx dy , (13.126)

−∞ −∞

where is the symbol of the Fourier transform. Introduce the variables u = x – λx and υ = y – λy . Then, Equation (13.126) takes the form ∞ ∞

[ a( x − λ x , y − λ y )] =

∫ ∫ a(u, υ) ⋅ e

−∞ −∞

= Sa (ω x , ω y ) ⋅ e

− j (ω xλ x + ω yλ y )

− j [ ω x ( u + λ x ) + ω y ( υ + λ y )]

du dυ (13.127)

.

Reference to Equation (13.127) shows us that only the phase component of the power spectral density of the object image has been changed.

Copyright 2005 by CRC Press

1598_book.fm Page 564 Monday, October 11, 2004 1:57 PM

564

Signal and Image Processing in Navigational Systems R (k) 1.00

0.75

1

2 3

0.25

4 k −6 −5 −4

−2 −1

0

1

2

3

4

6

FIGURE 13.12 The normalized criterial correlation functions: (1) L = 0; (2) L = 0.2; (3) L = 0.6; (4) L = 1.0.

The continuous narrowing of the criterial correlation function corresponds to the variations of the parameter L from zero to unity in Equation (13.120). If the object image has inﬁnite dimensions, the criterial correlation function is transformed into the delta function at L = 1: R( ∆λ) = −1 { e j∆ϕ } = δ ( ∆λ) .

(13.128)

The normalized criterial correlation functions for two square waveform signals shifted relative to each other by one discrete are shown in Figure 13.12. Reference to Figure 13.12 shows that at L = 1 the criterial correlation function has a sharp peak at the matching point of the object images. In narrowing the spatial correlation function, the steepness of the discrimination characteristic deﬁned by the derivative of the spatial correlation function of the moving X(x – λx) and model a*(x – λx* ) images is increased: D( ∆λ ) = −

dRa∗ ( ∆λ ) d ( ∆λ )

,

∆λ = λ x − λ∗x .

(13.129)

It is well known that estimations of the mutual positions of the moving and model images and the steepness of the discrimination characteristic are inversely proportional to each other. Therefore, the dependence of the steepness of the discrimination characteristic on the parameter L is of great interest. Let us assume that the mutual power spectral density of object images |Sa(ωx)|2 takes a bell-shaped form:27 2 2

|S a (ω x )|2 = e −0.5k ωx . Copyright 2005 by CRC Press

(13.130)

1598_book.fm Page 565 Monday, October 11, 2004 1:57 PM

Implementation Methods of the Generalized Approach to Space-Time

565

In this case, the criterial correlation function has the form −

R( ∆λ) = {e − 0.5 k

2

(1 − L )ω 2x

}=

e

∆λ 2 2 k 2 (1 − L )

.

k 2 π(1 − L)

(13.131)

Substituting Equation (13.131) in Equation (13.129), we obtain   − ∆λ − ∆λ d  e 2 k (1 − L )  ∆λ ⋅ e 2 k (1 − L ) . D( ∆λ) = − =  d( ∆λ)  k 2 π(1 − L)  k 3 2 π(1 − L)3   2

2

2

2

(13.132)

The ratio between the steepness of the discrimination characteristics at the condition 0 ≤ L ≤ 1 (the generalized image processing algorithm) and the condition L = 0 (the classical generalized processing algorithm) takes the following form: Q=

1 ( 1 − L) 3

.

(13.133)

As shown in Vasilenko,28 if the parameter L is varied from 0.2 to 0.4, the variance of the estimation error of mutual position between the moving and model images can be decreased by 1.5 to 2 times.

13.6 The Generalized Image Processing Algorithm: Invariant Moments The detection of the invariable peculiarities of object images is a very important landmark in pattern recognition theory. The invariable features of object images allow us to identify the object image independently of its location, dimensions, and orientation. The theory of algebraic invariants is the basis to detect the invariable or invariant peculiarities of object images. The theory of algebraic invariants studies a class of algebraic functions that do not vary under the transformations of coordinates. The ﬁrst attempt to use invariant moments with the pattern recognition of two-dimensional object images is discussed in Hu.29 The discussed technique is widely used in the recognition of object images on photos, in the recognition of aircraft in optical and radar range, and in searching moving clouds by satellite. The use of invariant moments in generalized image processing algorithm is preferred for the case when the image is the totality of the object and background, and therefore the transition to the background should be sufﬁciently sharp.

Copyright 2005 by CRC Press

1598_book.fm Page 566 Monday, October 11, 2004 1:57 PM

566

Signal and Image Processing in Navigational Systems

The central moments of the third order take the following form according to Equation (12.143):           

µ 00 = m 00 ; µ 11 = m 11 − Xm 10 ; µ 10 = 0 ; µ 30 = m 30 − 3am 20 + 2m 10 a 2 ; µ 01 = 0 ; µ 12 = m 12 − 2Xm 11 − am 02 + 2X 2 m 01 ;

(13.134)

µ 20 = m 20 − am 10 ; µ 21 = m 21 − 2am 11 − Xm 20 = 2a 2 m 01 ; µ 02 = m 02 − Xm 01 ; µ 03 = m 03 − 3Xm 02 + 2X 2 m 01.

The central moments are invariant with respect to the shift of object images. The normalized central moments can be determined in the following form: ν pq =

µ pq y µ 00

, where y = 0.5( p + q) and p + q = 2 , 3, … , . (13.135)

Using the central moments of the second and third order, we can generate a system consisting of seven invariant moments:30                   

ϕ1 = ν20 + ν02 ; 2 ϕ 2 = ( ν20 + ν02 )2 + 4ν11 ;

ϕ 3 = ( ν30 − 3 ν12 )2 + (3 ν21 + ν03 )2 ; ϕ 4 = ( ν30 + ν12 )2 + ( ν21 + ν03 )2 ; ϕ 5 = ( ν30 − 3 ν12 )( ν30 + ν12 )[(ν30 + ν12 )2 − 3( ν21 + ν03 )2 ] + (3 ν21 − ν03 )( ν21 + ν03 )[3( ν30 + ν21 )2 − ( ν21 + ν03 )2 ]; ϕ 6 = ( ν20 − ν02 )[( ν30 + ν12 )2 − ( ν21 + ν03 )2 ] + 4ν11 (ν30 + ν12 )(ν21 + ν 03 ); ϕ 7 = (3 ν12 − ν30 )( ν30 + ν12 )[(ν30 + ν12 )2 − 3(ν21 + ν03 )2 ] + (3 ν21 − ν03 )( ν21 + ν03 )[3( ν30 + ν12 )2 − ( ν21 + ν03 )2 ]. (13.136)

This system of moments is invariant with respect to the shift, rotation up to 45°, and double change of image scale.29,30 The criterial correlation function can be determined in the following form:

Copyright 2005 by CRC Press

1598_book.fm Page 567 Monday, October 11, 2004 1:57 PM

Implementation Methods of the Generalized Approach to Space-Time

Σ M N (∆x , ∆y )

567

7

i

i =1

R( ∆x , ∆y ) =

i

ΣM Σ N 7

7

i =1

2 i

i =1

2 i

,

(13.137)

(x , y )

where Mi is the i-th moment of the ﬁrst object image, and Ni is the i-th moment of the second object image shifted by the value (∆x, ∆y). The probability of true location of the object image with reference to a control point can be determined by

Ptr =

∞

e 2π σ ∫ 1

−

[ R ( x , y ) − R ( x *, y *)]2 2 σ2

dR( x , y),,

(13.138)

Kg

where R(x*, y*) is the maximum of the criterial correlation function for all points, Kg is the threshold, and σ2 is the variance of the probability distribution density of the criterial correlation function R(x, y). The probability of the false location of the object image with reference to the control point takes the following form:

Pfr =

Kg

e 2π R ∫ 1

−

[ R ( x , y ) − R ]2 2 R2

dR( x , y),

(13.139)

0

where R is the average of the criterial correlation function for all points.

13.7 The Generalized Image Processing Algorithm: Amplitude Ranking The criterial correlation function of rank form and hierarchical preprocessing are used in the implementation of the generalized image processing algorithm with amplitude ranking based on the generalized approach to signal processing in the presence of noise. The algorithm with amplitude ranking is used when there is a very large region for searching and there are high requirements for calculation accuracy and efﬁciency. The algorithm is a totality of algorithms, the complexity and calculation accuracy and efﬁciency of which can be optimized for deﬁnite parameters of the object image, for example, the dimensions of the object image, searching regions, etc. Let us consider the binary generalized image processing algorithm with amplitude ranking based on the generalized approach to signal processing Copyright 2005 by CRC Press

1598_book.fm Page 568 Monday, October 11, 2004 1:57 PM

568

Signal and Image Processing in Navigational Systems

in the presence of noise, which was constructed using the procedure discussed in Ormsby.31 This algorithm consists of two steps. The preliminary image processing is carried out at the ﬁrst step, in the course of which the smallest images are coded to binary correlation matrices. Therefore, we assume that either the moving image or the model image is much less in dimensions. These binary matrices are correlated sequentially with the decreasing set of the matrix having the larger dimension during the second step.32–34 The purpose of image preprocessing is to distribute the image elements in the order of the decreasing brightness characteristic and assigning the binary rank for each image element. This ranking procedure is shown in Figure 13.13 and Table 13.2. If the object image has 2D resolution elements, then the object image element with the lowest brightness characteristic takes the rank consisting of D zero and the object image element with the highest brightness characteristic takes the rank consisting of D units. Other object 1

2

3

1

11

17

1

2

6

5

15

3

14

4

10

Image 3 × 3 FIGURE 13.13 Ranking procedure.

TABLE 13.2 Ranking Procedure

Copyright 2005 by CRC Press

i, j

Brightness Characteristics

Binary Rank

1,2 2,3 3,1 1,1 3,3 2,1 2,2 3,2 1,3

17 15 14 11 10 6 5 4 1

111 110 101 100 XXX 011 010 001 000

1598_book.fm Page 569 Monday, October 11, 2004 1:57 PM

Implementation Methods of the Generalized Approach to Space-Time

569

image elements take a rank that is an element binary number in the ordered list. After ordering of the brightness characteristics, the object-image element list is divided into two different classes. Those from the ﬁrst class (object image elements with high brightness characteristics) take 1 and those from the second class take 0. If the number of the object image elements in a class is odd, then the object image element having a middle brightness characteristic (between 1 and 0) takes the index X. This procedure is iterated for each class of object image elements. Those with the index X keep this index until the procedure stops. The procedure stops when there is only one object image element in each class. After this procedure, the correlation matrices are formed. The ﬁrst correlation matrix C1 is formed by mapping the ﬁrst signs of the binary rank onto the initial object image. The D correlation matrices are formed in an analogous way (see Figure 13.14). The generalized image processing algorithm with amplitude ranking based on the generalized approach to signal processing in the presence of noise begins with the processing of the ﬁrst correlation matrix C1 and the largest object image L using the following technique: N ,M

N ,M

B1 ( x , y) =

∑

∑

Lx + i , y + j −

Lx + i , y + j ,

(13.140)

i , j = 1|C1 ( i , j ) = 0

i , j = 1|C1 ( i , j ) = 1

where N × M is the dimension of the object image. Thus, for each position of the matrix C1 on the largest object image, the sum of the object image elements at points corresponding to 0 in the matrix C1 is subtracted from the sum of the object image elements at points corresponding to 1 in the matrix C1. The object image elements at points corresponding to X are not taken into consideration. The function B1(x, y) is the primary correlation surface and characterizes the process of matching between the largest object image L and approximate structural information about the smallest object image. During this step, the points for which the value of B1(x, y) is low are rejected. For this purpose, a deﬁnite threshold is established. The secondary correlation surface B2(x, y), which is a more precise deﬁnition of the ﬁrst correlation surface B1(x, y), is determined by

C1 =

1

1

0

0

0

1

1

0

X

C2 =

FIGURE 13.14 Formation of N correlation matrices.

Copyright 2005 by CRC Press

0

1

0

1

1

1

0

0

X

C3 =

0

1

0

1

0

0

1

1

X

1598_book.fm Page 570 Monday, October 11, 2004 1:57 PM

570

Signal and Image Processing in Navigational Systems

N ,M

{ ∑

B2 ( x , y) = B1 ( x , y) + 0.5

N ,M

Lx + i , y + j −

i , j = 1|C1 ( i , j ) = 1

∑

}

Lx + i , y + j . (13.141)

i , j = 1|C1 ( i , j ) = 0

Calculations are carried out only for the object image elements, the brightness characteristics of which exceed the predetermined threshold. The described procedure is carried out for all correlation matrices and, in the general case, can be determined in the following form: N ,M

Bn ( x , y) = Bn−1 ( x , y) + 2

{ ∑

− ( n − 1)

i , j = 1|Cn ( i , j ) = 1

N ,M

Lx + i , y + j −

∑

}

Lx + i , y + j .

i , j = 1|Cn ( i , j ) = 0

(13.142) After determination of the last correlation surface, the point with the maximal magnitude of B1(x, y) is determined. This point is the point of matching between the moving image and the model image. The computational costs of the generalized image processing algorithm based on the generalized approach to signal processing in the presence of noise are characterized by the number of summation operations and determined by Q = (N – n) · (M – m) · nm,

(13.143)

where is the coefﬁcient 1 < < 2, N × M is the dimension of the largest object image, and n × m is the dimension of the smallest object image. The generalized image processing algorithm is an analog of the classical generalized signal processing algorithm. The performance and the probabilities of true and false location of the object image with reference to the control point are the same in practice.

13.8 The Generalized Image Processing Algorithm: Gradient Vector Sums The generalized image processing algorithm with gradient vector sums based on the generalized approach to signal processing in the presence of noise is invariant to rotation of the object image for the wide-angle range. During the ﬁrst step, the grey color gradient of the moving image and the model image are determined. During the second step, the histogram of the gradient vector sums as a function of the angle is formed for each object image if the gradient vectors are within the limits of discrete intervals of

Copyright 2005 by CRC Press

1598_book.fm Page 571 Monday, October 11, 2004 1:57 PM

Implementation Methods of the Generalized Approach to Space-Time

571

angles. Thereafter, these functions are processed either by the classical correlation function or by the criterial correlation function. The generalized image processing algorithm with gradient vector sums is based on the following principle. If the moving or model image is rotated by a deﬁnite angle, then the gradient vector determined for corresponding points of the moving and model images is rotated by the same angle. The brightness characteristics of corresponding points of the moving and model images are determined in the spherical coordinate system in the following form: G (ρ, θ ) = H (ρ′ , θ ′ ).

(13.144)

If, for instance, the model image is rotated with respect to the moving image by the angle ϕ relative to the coordinate system, then ρ = ρ′ and θ′ = θ + ϕ. In this case, the gradients at the corresponding points have the same value of ρ and are shifted with respect to each other by the angle determined by H (ρ, θ ) = G (ρ, θ ′ ) ⋅ e jϕ .

(13.145)

Because this condition is true for each pair of points, it is true for average values. For instance, if the average values of the object image gradients are determined in the following form 1 ∇Gav = 2 πρ 0

1 ∇H av = 2 πρ 0

2 π ρ0

∫ ∫ ∇G(ρ, θ) dρ dθ;

(13.146)

0 0

2 π ρ0

∫ ∫ ∇H (ρ, θ) dρ dθ,

(13.147)

0 0

where object images are deﬁned on the distance ρ0 from the origin of the coordinate system, then we can consider that H av = Gav ⋅ e jϕ ,

(13.148)

H (ρ, θ ) = G (ρ, θ − ϕ) ⋅ e jϕ .

(13.149)

because

Copyright 2005 by CRC Press

1598_book.fm Page 572 Monday, October 11, 2004 1:57 PM

572

Signal and Image Processing in Navigational Systems

Consequently, we can deﬁne without any difﬁculty the angle of relative shift of these object images ϕ = α1 – α2 , determining the average magnitude of gradients for two object images Gav = 1 ⋅ e iα1

and

H av = 2 ⋅ e iα 2 .

(13.150)

The main disadvantage of the generalized image processing algorithm with gradient vector sums is that the values of the average gradient are so close to zero that it makes this algorithm very critical with respect to errors caused by variations of low spatial frequencies. Therefore, the vector function determined by 2π

S(β) =

∫ ∫ ∇Gδ (θ − β) dS

ar

dθ

(13.151)

0 Sar

can be an alternative, where Sar is the area of object image. The vector function S(β) is deﬁned within the limits of the interval and can be considered as the sum of all gradient vectors of the object image having the angle β. In digital image processing, the gradient can be approximated by ﬁnite differences

∇G ( x m , y m ) ≡

∂G ˆ ∂G ˆ i+ j, ∂x m ∂y m

(13.152)

where ∂G = Gm + 2,n + 2Gm + 2,n+ 1 + Gm + 2,n+ 2 − Gm ,n − 2Gm ,n+ 1 − Gm ,n+ 2 ; (13.153) ∂x m ∂G = Gm ,n+ 2 + 2Gm + 1,n+ 2 + Gm + 2,n+ 2 − Gm ,n − 2Gm + 1,n − Gm + 2,n . (13.154) ∂y m Experimental results conﬁrming the possibility of identifying two object images rotated relative to each other by angles of up to 30° are discussed in Davies and Bouldin.35

Copyright 2005 by CRC Press

1598_book.fm Page 573 Monday, October 11, 2004 1:57 PM

Implementation Methods of the Generalized Approach to Space-Time

573

13.9 The Generalized Image Processing Algorithm: Bipartite Functions Let us consider the following criterial bipartite correlation functions:36,37 2n − 1

R1 (u, υ) =

∏T ;

(13.155)

i

i=0

2n − 1 2n − 1

R2 (u, υ) =

Σ Σ (i, j)F (u, υ) i=0 j=0

2n − 1

2n − 1

ij

2n − 1

.

2n − 1

(13.156)

Σ i [ Σ F (u, υ)] Σ j [ Σ F (u, υ)] 2

2

j=0

i=0

ij

j=0

i=0

ij

The bipartite correlation function R1(u, υ) given by Equation (13.155) is a weighted product and depends on the number of quantization levels and weighted estimation of coincidences and noncoincidences of the brightness characteristic levels. For example, for two and four quantization levels we can write T0 =

Copyright 2005 by CRC Press

F00 R 0 + F01

and T1 =

F11 ; R 1 + F10

(13.157)

T0 =

2F00 + F01 ; 2R0 + F02 + 2F03

(13.158)

T1 =

2F11 + F10 + F12 ; 2R1 + F13

(13.159)

T2 =

2F22 + F21 + F23 ; 2R2 + F20

(13.160)

T3 =

2F33 + F32 , 2R3 + F31 + 2F30

(13.161)

1598_book.fm Page 574 Monday, October 11, 2004 1:57 PM

574

Signal and Image Processing in Navigational Systems 2n − 1

where Ri =

Σ F is the number i for the model image. Thus, noncoincidences

j=0

ij

for one quantization level can be considered as coincidences for another. The bipartite criterial function R2(u, υ) is very close to the classical generalized signal processing algorithm based on the generalized approach to signal processing in the presence of noise. The bipartite criterial correlation functions given by Equation (13.22) and Equation (13.23) are the products of ratios of the number of object image elements with the same brightness characteristics and the number of possible coincidences of all forms of the brightness characteristics. Note that the number of bipartite functions increases as the square of the number of quantization levels. Consequently, we should try to reduce the number of quantization levels. In binary image processing, there are four forms of the bipartite function united into a matrix F00 F01 F10 F11

.

(13.162)

If the object images are identical, then only the functions on the main diagonal of the matrix given by Equation (13.162) are nonzero functions. The binary object images can be deﬁned by four bipartite functions. In this case, based on Equation (13.22) and Equation (13.23), we can write R( u, υ) =

F00 F11 ⋅ . F00 + F01 F11 + F10

(13.163)

The criterial function given by Equation (13.163) has a narrow main lobe and the level of the side lobes is also more narrow.

13.10 The Hierarchical Generalized Image Processing Algorithm The basis of the hierarchical generalized image processing algorithm is a procedure in image preprocessing in which a set of object images with sequentially decreasing resolutions is formed from the initial object images. Various criterial functions are used to compare the obtained sets of object images. The hierarchical generalized image processing algorithm begins with the comparison of object images having the lowest resolution level. In image processing, a set of most probable control points is chosen at each Copyright 2005 by CRC Press

1598_book.fm Page 575 Monday, October 11, 2004 1:57 PM

Implementation Methods of the Generalized Approach to Space-Time

575

level. At the next levels, computer calculations are carried out only for those points that were chosen at the previous levels. Therefore, the number of processed points is reduced sharply, which ensures the high effectiveness of the algorithm. In the hierarchy generalized image processing algorithm, the number of processed points at the k-th level is determined by nk =

[ N −2M + 1 + 1] . 2

(13.164)

L

The value of nk is 22L times less than the number of processed points of the object image with an initial high resolution. The decrease in the number of processed points obeys the logarithmic law B log (N – M + 1)2, where B is a constant. As a rule, the object image Xk-1(x, y) at the (k – 1)-th level is ﬁltered by the linear low-frequency ﬁlter h(x, y) with the transfer function C(ωx , ωy): X 0 ( x , y ) = X k−1 ( x , y ) ∗ h ( x , y ).

(13.165)

The digitization function takes the following form: ∞

d( x , y) =

∞

∑ ∑ δ(x − 2j ∆x, y − 2j ∆y), 1

(13.166)

2

j2 = − ∞ j1 = − ∞

where ∆x and ∆y are the intervals of digitization at the (k – 1)-th level. The Fourier transform at the k-th level after digitization is determined by

Sk (ω x , ω y ) =

π2 ∆x∆y j

∞

∞

∑ ∑S

k −1

(ω x − j∆πx , ω y − j∆yπ ) ⋅ C(ω x − j∆πx , ω y − j∆yπ ). 1

2

1

2

1 = − ∞ j2 = − ∞

(13.167) The transfer function of the low-frequency ﬁlter can be written in the following form: C (ω x , ω y ) = [cos( 0.5ω x ) ⋅ cos( 0.5ω y )]n ,

(13.168)

where n characterizes the degree of signal damping by the high-frequency ﬁlter. According to Hall,18 sufﬁcient signal damping is ensured at n = 8. In digital signal processing, we assume that Xk-1(n1 , n2) is the moving image at the ﬁlter input, and C(l1 , l2) is the transfer function of the ﬁlter. The object image at the k-th level at the ﬁlter output can be determined in the following form: Copyright 2005 by CRC Press

1598_book.fm Page 576 Monday, October 11, 2004 1:57 PM

576

Signal and Image Processing in Navigational Systems m1 − l1 + 1

X k (m1 , m2 ) =

m2 − l2 + 1

∑ ∑

n1 = m1

X k −1 (n1 , n2 ) ⋅ C(n1 − m1 + 1, n2 − m2 + 1).

n2 = m2

(13.169) At n = 8, we obtain 1 1 1 C ( l1 , l2 ) =

1 ⋅ 1 2 1 10 1 1 1

.

(13.170)

Let us consider the two-level hierarchy generalized image processing algorithm. In the initial step, the processing of the moving image and the part with the most informative features of the model image is carried out with all possible relative shifts of images. During the second step, the total model image is used, but a comparison is carried out only at the points with maximal correlation that were chosen during the ﬁrst step. Computational costs J in using the hierarchy generalized image processing algorithm can be estimated in the following form: J = A + Pthr V ,

(13.171)

where A is a computational cost during the ﬁrst step; Pthr is the probability of exceeding the threshold during the ﬁrst step, that is, the probability of deﬁnition of the point that will be processed during the second step and V is the computational cost in using the total model image during the second step of the hierarchy generalized image processing algorithm. In the general case, the greatest part of the model image is used during the ﬁrst step and the computational cost is high, but, as a result, the probability Pthr is decreased. Let {t1, … , tn} be the values of the grey level of n points of the model image and {g1, … , gn} the values of the grey level of the moving image part arbitrarily chosen. The criterial correlation function can be written in the following form: n

R=

∑| g − t |. i

i

(13.172)

i=1

In this case, the computational cost is proportional to the dimensions of the model image if the grey level of background obeys the Gaussian probability distribution density and the absolute values of the brightness characteristics of the moving and model images are independent. Copyright 2005 by CRC Press

1598_book.fm Page 577 Monday, October 11, 2004 1:57 PM

Implementation Methods of the Generalized Approach to Space-Time

577

13.11 The Generalized Image Processing Algorithm: The Use of the Most Informative Area In this case, the peculiarity of the generalized image processing algorithm is the image preprocessing in the course of which the most informative areas are detected.38 This technique ensures a decrease in computational cost. Object image elements, the brightness characteristics of which are high compared to the average value of the brightness characteristics, can be used as the most informative areas. To choose the most informative areas, it is necessary to obtain the probability distribution density of the object image elements as a function of the brightness characteristic (x) (see Figure 13.15), where m and σ are the mean and the variance of the brightness characteristic (x), respectively, and ρ ≥ 0 is a constant. The object image g′(i, j) is formed as a result of preprocessing the object image g(i, j). The object image g′(i, j) contains the elements of the initial object image in accordance with the formula g ′(i, j) ∈{ (m − ρσ ) ≥ g(i, j) ≥ (m + ρσ )} .

(13.173)

Thus, the volume of reduced information depends on the behavior of the probability distribution density of the brightness characteristics (x) and the parameter ρ. For example, if the brightness characteristic (x) obeys the Gaussian probability distribution density

−1

f ( x) = ( 2 π σ ) ⋅ e

−

( x − m )2 2 σ2

,

(13.174)

f (N)

L(x) m − ρσ

m

m + ρσ

FIGURE 13.15 Numbers of object image elements as a function of the brightness characteristics.

Copyright 2005 by CRC Press

1598_book.fm Page 578 Monday, October 11, 2004 1:57 PM

578

Signal and Image Processing in Navigational Systems

then the ratio of the element numbers of the object image g′(i, j) to the element numbers of the object image g(i, j) can be determined in the following form: m + ρσ

β = 1−

∫

m − ρσ

2 f ( x) dx = 1 − π

ρ

∫e

− 0.5 x 2

dx .

(13.175)

0

Thus, the coefﬁcient of the computational cost reduction in comparison with the classical generalized image processing algorithm can be determined by γ=

ρ

e 2π ∫

200

− 0.5 x 2

(13.176)

dx.

0

The dependence of the coefﬁcient γ on the parameter ρ, if the condition 0 < ρ < 1 is satisﬁed, has a linear character. For example, at ρ = 1, we obtain γ = 65%. If the probability distribution density fL(x) can be presented as the totality of two Gaussian laws with the parameters m1, σ1 , m2 , and σ2 , then the coefﬁcient γ can be determined by

[ ( P (m −σm ) + ρσ ) − erf ( P (m

− m 1) + ρσ σ1

)]

P ( m − m 2 ) + ρσ P ( m − m 2 ) + ρσ + P2 erf 1 1 − erf 1 1 σ2 σ2

)]

γ = P1 erf

2

1

2

1

2

1

[ (

)

(

, (13.177)

where x

−1

erf ( x) = ( 2 π )

∫e

− 0.5 λ 2

dλ ;

(13.178)

0

σ 2 = P1P2 ( m 1 − m 2 ) 2 + P1σ 21 + P2σ 22 ,

(13.179)

P1 and P2 are the a priori probabilities of appearance of one of the two main levels of the object image brightness characteristics, and therefore, P1 + P2 = 1.

13.12 The Generalized Image Processing Algorithm: Coding of Images The main peculiarity of the generalized image processing algorithm in the coding of images based on the generalized approach to signal processing in the presence of noise is the preliminary image processing based on the use Copyright 2005 by CRC Press

1598_book.fm Page 579 Monday, October 11, 2004 1:57 PM

Implementation Methods of the Generalized Approach to Space-Time

579

of various techniques of economic coding and removal of the excessive informativeness of images. One of the methods that is used in the generalized image processing algorithm is cut block coding. This technique realizes principles of adaptive block coding applied to the adaptation of local information structures, not the average brightness characteristics of the object image. The essence of cut block coding is the following. It is necessary to divide the initial image X(i, j) into blocks consisting of n × m elements with the sequential binary quantization of each block at the levels a and b. It is recommended that the linear dimensions of blocks be 3 to 5 times less than the correlation length of the object image. There are some image preprocessing procedures that use the technique of choosing the levels and quantization threshold Kg in the generalized image processing algorithm with cut block d coding. The determination of the levels and quantization threshold can be carried out in the following manner:

a= x−σ

q , p−q

b= x+σ

p−q , q

K gd = x12 ,

and

(13.180)

where

x=

1 mn

n

m

∑ ∑ X(i, j);

(13.181)

i=1 j=1

σ 2 = x 12 − x ; 1 x = mn 2 1

n

(13.182)

m

∑ ∑ X (i, j); 2

(13.183)

i=1 j=1

x and σ2 are the moment and variance deﬁned for each block, p = mn, and q is the number of elements of the object image X(i, j) exceeding the threshold Kg . There is another technique, based on which the parameters of the quantizer are chosen to maximize the criterial correlation function. For this purpose, the elements of the object image block Xk(i, j) are ranked in increasing order. Based on the obtained sequence zk1 , … , zkp , we determine the quantization levels ak =

Copyright 2005 by CRC Press

1 p − qk

p − qk

∑ i=1

zki

and

bk =

1 qk

p

∑

i = p − qk + 1

zki ,

(13.184)

1598_book.fm Page 580 Monday, October 11, 2004 1:57 PM

580

Signal and Image Processing in Navigational Systems

where qk is the number of elements zki exceeding the threshold Kgd. The threshold value is deﬁned by the maximization of the criterial correlation function compared to corresponding blocks of the initial object image and the coding object image: p − qk

R = ak

∑z

p

ki

∑

+ bk

i=1

(13.185)

zki .

i = p − qk + 1

It is also possible to deﬁne simultaneously the quantization levels in accordance with Equation (13.180) and the maximum of the criterial correlation function given by Equation (13.185).

13.13 The Multichannel Generalized Image Processing Algorithm The multichannel generalized image processing algorithm is used to increase the accuracy of shift estimation between the moving and model images that could be made using channels added in parallel to the main channel of the generalized receiver. These additional channels of the generalized receiver process components of the moving and model images. Preliminary image processing is carried out by nonlinear ﬁlters with Π-form characteristics to obtain these components of the moving and model images. As is well known, the accuracy of shift estimation between the moving and model images is proportional to the steepness of the discrimination characteristic which, taking into consideration additional channels of the generalized receiver, can be determined in the following form:

{ dad R

DN (α) = −

( a) + aX ∗

d da

N

∑rR i

i=1

∗ a∗Xi

}

( a) ,

(13.186)

where Ra*X(a) is the correlation function between the moving and model images, ri are the coefﬁcients, N is the number of additional channels of the * generalized receiver, and Ra*X (a) is the correlation function between comi ponents of the moving and model images. If the brightness characteristics of these components obey the Gaussian probability distribution density, we * can write the correlation function Ra*X (a) in the following form: i ∞

Ra∗∗X ( a) = i

Copyright 2005 by CRC Press

∑ (k !) k=0

−1

hk2i Rak∗ ( a);

(13.187)

1598_book.fm Page 581 Monday, October 11, 2004 1:57 PM

Implementation Methods of the Generalized Approach to Space-Time

2

hki =

ai 2π σ k

{H

( )⋅e

bi k −1 σ

b − i

2 σ2

581

2

− H k −1 (

bi + ci σ

)⋅e

(b + c ) − i i 2 σ2

},

(13.188)

where Hn(.) is the Hermitian polynomial. In the only additional channel of the generalized receiver, the accuracy of shift estimation is increased 2.7 times. Thus, the use of the multichannel generalized image processing algorithm allows us to increase signiﬁcantly the accuracy of matching between the moving image and the model image.

13.14 Conclusions In this chapter, we consider brieﬂy some quasioptimal generalized image processing algorithms in the use of the various types of criterial correlation functions that are constructed based on of the generalized approach to signal processing in the presence of noise. Each quasioptimal generalized image processing algorithm has both advantages and disadvantages. We will discuss these advantages and disadvantages brieﬂy. The classical generalized image processing algorithm is highly efﬁcient with low values of signal-to-noise ratio but has a very high computational cost. The difference generalized image processing algorithm allows us to reduce computational costs signiﬁcantly and has better detection performance with high values of signal-to-noise ratio in comparison with the classical generalized image processing algorithm. With low values of the signal-to-noise ratio, the detection performance of the difference generalized image processing algorithm is less compared to the classical generalized image processing algorithm. The generalized image processing algorithm with bipartite functions displays a very high efﬁciency in the computational cost. However, the use of the sign functions in the implementation of the generalized image processing algorithm with the bipartite functions is possible only in the case of the Gaussian probability distribution density. The phase generalized image processing algorithm ensures very spiked peaks of the criterial correlation function, displays a high efﬁciency in the computational cost, and is nonsensitive to narrow-band noise. However, it is very highly sensitive to high-frequency distortions. The use of the hierarchy generalized image processing algorithm ensures a low computational cost, but it requires an increase in the volume of computer memory. Its signiﬁcant disadvantage is the very high complexity of detection of informative features. The generalized image processing algorithm with amplitude ranking has approximately the same detection performance as the classical generalized image processing algorithm. Low computational cost is realized only in the case when the object image has large dimensions. The generalized

Copyright 2005 by CRC Press

1598_book.fm Page 582 Monday, October 11, 2004 1:57 PM

582

Signal and Image Processing in Navigational Systems

image processing algorithm with invariant moments is invariant to rotation of object images, but its computational efﬁciency in image processing is low.

References 1. Wang, R. and Hall, E., Sequential hierarchical scene matching, IEEE Trans., Vol. C-27, No. 4, 1978, pp. 359–366. 2. Arsenault, H. et al., Incoherent method for rotation-invariant recognition, Appl. Opt., Vol. 21, No. 4, 1982, pp. 610–615. 3. K. Fu, Ed., Digital Pattern Recognition, Springer-Verlag, Berlin, 1976. 4. Knapp, C. and Carter, G., The generalized correlation method for estimation of time delay, IEEE Trans., Vol. ASSp-24, No. 4, 1976, pp. 320–327. 5. Novak, L., Correlation algorithms for radar map matching, IEEE Trans., Vol. AES-14, No. 4, 1978, pp. 641–648. 6. Mirskiy, G., Characteristics of Stochastic Correlation and Its Measurement, Energoizdat, Moscow, 1982 (in Russian). 7. Tuzlukov. V., A new approach to signal detection theory, Digital Signal Process. Rev. J., Vol. 8, No. 3, 1998, pp. 166–184. 8. Tuzlukov, V., Signal Processing in Noise: A New Methodology, IEC, Minsk, 1998. 9. Tuzlukov, V., Signal Detection Theory, Springer-Verlag, New York, 2001. 10. Tuzlukov, V., Signal Processing Noise, CRC Press, Boca Raton, FL, 2002. 11. Kuglin, C. et al. Map-matching techniques for terminal guidance using Fourier phase information, in Proceedings of the SPIE, Vol. 186, 1979, pp. 21–29. 12. Vladimirov, V., Equations in Mathematical Physics, Nauka, Moscow, 1971. (In Russian.) 13. Kuglin, C. and Hines, D., The phase correlation image alignment method, in Proceedings of the IEEE International Conference on Cybernetics and Society, 1975, pp. 163–165. 14. Ausherman, D., Kozma, A., Walker, J., Jones, H., and Poggio, E., Developments in radar imaging, IEEE Trans., Vol. AES-20, No. 3, 1984, pp. 363–279. 15. Stark, H. and Woods, J., Polar sampling theorems and their applications to computer-aided tomography, in Proceedings of the SPIE, Vol. 231, 1980, pp. 230–241. 16. Candel, S., Dual algorithms for fast calculation of the Fourier-Bessel transformations, IEEE Trans., Vol. ASSP-29, No. 5, 1981, pp. 963–972. 17. Herman, G., Image Reconstruction from Projections, Springer-Verlag, New York, 1979. 18. Hall, E., Computer Image Processing and Recognition, Academic Press, New York, 1979. 19. Budinger, T. and Gullbery, G., Three-dimensional reconstruction in nuclear medicine emission imaging, IEEE Trans., Vol. NS-21, No. 6, 1974, pp. 2–20. 20. Cook, G. et al., Optimal reconstruction angles, in Proceedings of the IEEE Conference on Computers in Radiology, 1979, pp. 291–304. 21. Clary, J. and Russell, R., All-digital correlation for missile guidance, in Proceedings of the SPIE, Vol. 119, 1977, pp. 36–46. 22. Stubberud, A., Acquisition probability for a correlation algorithm, in Proceedings of the IEEE Circuits and Systems Conference, 1981, pp. 263–265.

Copyright 2005 by CRC Press

1598_book.fm Page 583 Monday, October 11, 2004 1:57 PM

Implementation Methods of the Generalized Approach to Space-Time

583

23. Rockmore, A., The probability of false acquisition for image registration, Image Science Mathematics Symposium, 1976, pp. 252–255. 24. Xu, X. and Narayanan, R., Three-dimensional interferometric ISAR imaging for target scattering diagnosis and modeling, IEEE Trans., Vol. IP-10, No. 7, 2001, pp. 1094–1102. 25. Zebker, H. and Lu, Y., Phase unwrapping algorithms for radar interferometry: reside-cut, least-squares, and synthesis algorithms, J. Opt. Soc. Am., Vol. 15, No. 3, 1998, pp. 586–598. 26. Herrero, J., Portas, J., and Corredera, J., Use of map information for tracking targets on airport surface, IEEE Trans., Vol. AES-39, No. 2, 2003, pp. 675–693. 27. Johnson, M., Analytical development and test results of acquisition probability for terrain correlation devices used in navigation systems, AIAA, Paper N72–122. 28. Vasilenko, G., Holography Pattern Recognition, Soviet Radio, Moscow, 1977. (In Russian.) 29. Hu, M.-K., Visual pattern recognition by moment invariants, IRE Trans., Vol. IT-8, 1962, No. 1, pp. 179–187. 30. Wong, R. and Hall, E., Scene matching with invariant moments, Comput. Vision Grap., Vol. 8, No. 1, 1978, pp. 16–24. 31. Ormsby, C., Advanced scene matching techniques, in Proceedings of IEEE NAECON, 1979, pp. 68–78. 32. Bryant, M., Gostin, L., and Soumekh, M., 3-D E-CSAR imagimg of a T-72 tank and synthesis of its SAR reconstructions, IEEE Trans., Vol. AES-39, No. 1, 2003, pp. 211–227. 33. Collins, N. and Baird, C., Terrain aided passive estimation, in Proceedings of the IEEE National Aerospace and Electronics Conference, 1989, Vol. 3, pp. 909–916. 34. Nabaa, N. and Bishop, R., Validation and comparison of coordinated turn aircraft maneuver models, IEEE Trans., Vol. AES-36, No 1, 2000, pp. 250–259. 35. Davies, D. and Bouldin, D., Correlation of rotated images by the method of gradient vector sums, in Proceedings of IEEE SOUTHEASTCON, 1979, pp. 367–372. 36. Garett, G. et al., Detection threshold estimation for digital area correlation, IEEE Trans., Vol. SMC-6, No. 1, 1976, pp. 65–70. 37. Wong, R., Sequential scene matching using edge features, IEEE Trans., Vol. AES14, No. 1, 1978, pp. 128–140. 38. Ranganath, H. et al., Feature extraction technique for fast digital image registration, in Proceedings of IEEE SOUTHEASTCON, 1980, pp. 225–228.

Copyright 2005 by CRC Press

1598_book.fm Page 585 Monday, October 11, 2004 1:57 PM

14 Object Image Preprocessing

14.1 Object Image Distortions The quasioptimal generalized signal and image processing algorithms based on the generalized approach to signal processing in the presence of noise discussed in Chapter 13 are intended for the processing of identical object images having a deﬁnite relative shift. It is assumed that distortions are absent, which is not always true. In practice, images of the same object obtained at different times or by various sensors can be fundamentally different from each other. So, object image preprocessing is necessary to bring the images into one-to-one correspondence. The quality of the preprocessing will have an impact on the general parameters of any navigational system. Let us consider some distortions generated by electromagnetic ﬁelds. All object image distortions in form can be divided into geometrical distortions and distortions in brightness characteristics. Both types of distortions are caused by various reasons. For example, geometrical distortions of object images are possible due to inaccurate information about the object’s position in space. In this case, distortions for various spatial coordinates are shown in Figure 14.1. Geometrical distortions arise during transmission of object images by various sensors when an implementation of these sensors assumes various tracking angles. Distortions in brightness characteristics of object images arise, as a rule, due to changes in meteorological or seasonal conditions.

14.2 Geometrical Transformations Object images obtained by various sensors, for example, by radar sidescanning and optical scanning, can be brought into one-to-one correspondence with a geometrical transformation or correction. We consider two techniques that can be applied with the use of the generalized image

Copyright 2005 by CRC Press

1598_book.fm Page 586 Monday, October 11, 2004 1:57 PM

586

Signal and Image Processing in Navigational Systems

(a)

(b)

(d)

(c)

(e)

FIGURE 14.1 Distortions for various spatial coordinates: (a) altitude; (b) pitch; (c) complex error; (d) heading; (e) displacement.

processing algorithm based on the generalized approach to signal processing in the presence of noise: the perspective transformation and polynomial estimation.1

14.2.1 The Perspective Transformation The principal scheme of object image generation, which is also of general applicability, uses two sensors, with geometrical features of optical systems and radar side-scanning included in the navigational system, and is shown in Figure 14.2. The image plane of an optical system and the object plane (the Earth’s surface) are parallel (see Figure 14.3). The optical axis is orthogonal to both planes. Because the distance between the optical system and the object is much greater than the focus f, the image plane and the focal plane coincide and x3 = f. Each point of the object image u′ = (u1′ , u2′ , u3′ ) has a corresponding point on the focal plane x = ( x1 , x2 , x3 ). The components of the point x can be determined as follows x1 = f ⋅

u1′ , u3′

x2 = f ⋅

u2′ , u3′

and

x3 = f .

(14.1)

The object image obtained by radar side-scanning (see Figure 14.4) depends on the angles of pitch θ, roll γ, and hunting ψ, which deﬁne the position of the navigational system. The object point u = (u1, u2, u3) on the radar side-

Copyright 2005 by CRC Press

1598_book.fm Page 587 Monday, October 11, 2004 1:57 PM

Object Image Preprocessing

587 Optical System

Radar

FIGURE 14.2 Optical system and radar side-scanning.

3

2

(x1, x2) Image Plane

x3

u ′3

1 u ′1

Optical Axis

u ′2

The Earth Surface

FIGURE 14.3 Optical system and the Earth’s surface.

scanning plane can be deﬁned as the point u′ = (u1′ , u2′ , u3′ ) in the coordinate system of the optical sensor by a formula in matrix form u′ = M ⋅ u

Copyright 2005 by CRC Press

(14.2)

1598_book.fm Page 588 Monday, October 11, 2004 1:57 PM

588

Signal and Image Processing in Navigational Systems 3 2′ 3′

x2

2

x1 1′

Image Plane

x3

f u3

ψ

θ

u1 u2

1 Optical Axis The Earth Surface

FIGURE 14.4 Radar side-scanning object image.

where M is the matrix of rotation, the elements of which are functions of the angles θ, γ, and ψ.1 Substituting Equation (14.2) in Equation (14.1), we get x1 = f ⋅

m11u1 + m12 u2 + m13 u3 ; m31u1 + m32 u2 + m33 u3

(14.3)

x2 = f ⋅

m21u1 + m22 u2 + m23 u3 ; m31u1 + m32 u2 + m33 u3

(14.4)

x3 = f ,

(14.5)

where mij are the elements of the rotation matrix M. Using Equation (14.3) and Equation (14.4), the object image point with coordinates (u1, u2) can be transformed into the coordinates (x1, x2) of the optical system. Note that the transforms given by Equation (14.3)–Equation (14.5) are characteristic of object images obtained by optical and heat vision apparatus. Radar side-scanning represents the object image in the following coordinates: sloped distance and azimuth. So, it is necessary to ﬁrst transform the object image in the radar-side-scanning plane to the coordinate system of the angles θ, γ, and ψ to use this technique.

Copyright 2005 by CRC Press

1598_book.fm Page 589 Monday, October 11, 2004 1:57 PM

Object Image Preprocessing

589

14.2.2 Polynomial Estimation Let f (u1, u2) be the radar image and f (x1, x2) be the optical image. The polynomial approach is to use the following formulae:2 N

N −i

∑∑ a u u ;

x1 =

i j ij 1 2

(14.6)

i=0 j=0

N

x2 =

N −i

∑∑b u u , i j ij 1 2

(14.7)

i=0 j=0

where aij and bij are the constant polynomial coefﬁcients. In practice, we use polynomials of the second order, i.e., N = 2, as a rule. To obtain coefﬁcients it is necessary to choose the most signiﬁcant features of the radar and optical object images, such as the ﬁnite points of long lines, of intersections, and so on. With the use of a polynomial of the second order, it is necessary to choose no less than six conjugate points. The values of these points are as follows:

x11 x12 x13

a00

1 u1 u2 u12 u22 u1 u2

a10

=

x1n

⋅

a01

,

(14.8)

a20 a02 a11

where n ≥ 6. The analogous formulae can be deﬁned for x2 and bij and can be expressed in matrix form: x1 = u · A and x2 = u · B. Estimates A and B are deﬁned by the following equations: A = (u T u)−1 u T x 1

and B = (u T u)−1 u T x 2 .

(14.9)

The polynomial estimation technique is preferred because additional information is not required; information included in the obtained object images is sufﬁcient. Consider the method of preprocessing active sensor signals by measurement of radar range with a radar or a scanning laser system.3 Sensor data can be presented in spherical coordinates: radar range and two angles. Because the generalized signal and image processing algorithms are in the

Copyright 2005 by CRC Press

1598_book.fm Page 590 Monday, October 11, 2004 1:57 PM

590

Signal and Image Processing in Navigational Systems x

α α

_β

β

_β

x1 x2

y1 = y2 α

y

R

z2 z = z1 FIGURE 14.5 Coordinate systems.

Cartesian coordinate system, it is necessary to transform the object image to the Cartesian coordinate system and carry out additional processing to reduce geometrical distortions during the preprocessing step. These coordinate systems are shown in Figure 14.5. Here, R = R ⋅ x2 , where R is the radarrange vector x1

x

y

= A⋅ y

z

z

;

(14.10)

cos α sin α 0 A = − sin α cosα 0 0 x2

0

;

1

x1

x

y1 = B ⋅ y1 = B ⋅ A ⋅ y z2

(14.11)

;

(14.12)

z

z1 cos β 0 sin β

B=

0

1

0

.

− sin β 0 cos β The coordinate transformation takes the following form:

Copyright 2005 by CRC Press

(14.13)

1598_book.fm Page 591 Monday, October 11, 2004 1:57 PM

Object Image Preprocessing

591

x y

R −1

=A B

−1

,

0

z

(14.14)

0

where A–1 and B–1 are the matrices obtained by replacing the variables α and β with – α and – β in the matrices A and B, respectively. The matrices A–1 and B–1 are the reciprocal matrices. Coordinate transformation does not inﬂuence the brightness characteristic I for any value of the matrix R. The vector of parameters I, R, α, and β is transformed into the vector of parameters I, x, y, and z, respectively. The second step of image preprocessing is to obtain a projection of the object image on the plane that is orthogonal to the tracking line. This transformation is carried out by two sequential rotations of the Cartesian coordinates (x, y, z): the ﬁrst rotation by the angle αp with respect to the axis z and the second rotation by the angle β with respect to the axis y. The projection plane coordinates are xp , yp , and zp . Because of this, we can write xp

x

yp = Bp Ap

y

zp

z

,

(14.15)

where Ap and Bp are the matrices obtained by replacing the variables α and β with αp and βp in the matrices A and B, respectively. Taking into account Equation (14.13), we can write xp

R −1

yp = Bp Ap A B zp

−1

0

.

(14.16)

0

The values of R, A–1, and B–1 vary from point to point of the sensor searching area, whereas the values of Ap and Bp are constant. As the radar range vector R lies along the axis x2 (see Figure 14.5), it is necessary to determine the elements of the ﬁrst column of the matrix BpAp A–1B–1. The second step of the transformation is illustrated by Figure 14.6, in which the point (xp , yp , zp) in the projection plane corresponds to the point (R, α, β) in the spherical coordinate system.

Copyright 2005 by CRC Press

1598_book.fm Page 592 Monday, October 11, 2004 1:57 PM

592

Signal and Image Processing in Navigational Systems

Sensor

0, yp , zp

xp yp zp

R, α, β

Projection Plane

xp, yp , zp

Sensor Searching Area FIGURE 14.6 The second step of the coordinate transformation.

14.2.3 Transformations of Brightness Characteristics Images of the same objects obtained by various sensors have different distributions of brightness characteristics. If, for example, the optical object image is positive, then the radar image of the same object is negative. An example of the brightness characteristics distribution of optical and radar images of the same object is shown in Figure 14.7 and Figure 14.8. The generalized signal and image processing algorithm for these images cannot bring positive results. The object images must be brought into one-to-one correspondence. Let us consider two transformation techniques of object image brightness characteristics that are discussed in References 2 and 4.2,4 The main feature of the ﬁrst technique is the use of statistical characteristics of object images. The second technique does not require this information, because the radar object image is transformed so that the brightness characteristics correspond to the exponential distribution law. The ﬁrst transformation step of brightness characteristics is a transition from the negative radar image to the positive radar image. This transformation is carried out by replacing the brightness characteristic of each image element by the brightness level eij. Then, eij = 2 n − eij − 1 ,

Copyright 2005 by CRC Press

(14.17)

1598_book.fm Page 593 Monday, October 11, 2004 1:57 PM

Object Image Preprocessing

593

Brightness

Grey level 0

50

100

150

200

250

FIGURE 14.7 The brightness characteristic distribution of an optical object image.

Brightness

0

50

100

150

200

250 Grey level

FIGURE 14.8 The brightness characteristic distribution of a radar object image.

where 2n is the number of quantization levels. Assume that the radar image R and the optical image P have the dimensions N × N: R = (r1 , ... , rN × N )

Copyright 2005 by CRC Press

and

P = ( p1 , ... , pN × N ) .

(14.18)

1598_book.fm Page 594 Monday, October 11, 2004 1:57 PM

594

Signal and Image Processing in Navigational Systems

We investigate each element pair (rk , pk). Introduce the bipartite function F(y1, y2), where y1 is the brightness characteristic level of the element rk and y2 is the brightness characteristic level of the element pk . Under the condition y1 = y2 the function F(y1, y2) is equal to the number of image elements having the same brightness characteristics. Under the condition y1 ≠ y2 , the function F(y1, y2) deﬁnes the number of noncoinciding image elements. The covariance matrix takes the form

{

Cy = M

y − My ⋅ y − My

T

},

(14.19)

where y1

y=

My =

and

y2

M y1

,

(14.20)

M y2

My1 and My2 are the expected levels y1 and y2 of brightness characteristics, respectively; M is the symbol of the mean. The radar image is transformed in the following manner. The number of image elements of each brightness characteristic level coincides with the corresponding number of optical image elements. For this purpose, we introduce a new coordinate system g1 and g2 so that Cλ = GCyG–T, where G is the matrix whose columns are the eigenvectors Cy with the eigenvalues λ1 and λ2. The required correction of brightness characteristics is reached when the angle between g1 and y1 becomes approximately equal to 45°. With the second technique, the radar object image is transformed into a positive form [see Equation (14.6)]. A further transformation replaces the brightness characteristics eij′ so that N0 ≤ eij ≤ N k and M0 ≤ eij′ ≤ ML . The transformation is carried out so that the probability distribution law at the output P[ eij′ = M ], 0 ≤ ≤ L takes an expected shape at the given probability distribution law P[ eij = Nk], 0 ≤ k ≤ K. The transformation takes the form:

k

∑ q= 0

P[eij = Nq ] =

∑ P[e′ = M ] . ij

k

(14.21)

x=0

To obtain the uniform probability distribution law P[ eij′ ] = ( ML − M0 )−1 , the transformation should take the following form eij′ = [ ML − M0 ] ⋅ P [ eij ] + M0 .

(14.22)

To obtain the Rayleigh law with the parameter β > 0, the following equality must be satisﬁed:

Copyright 2005 by CRC Press

1598_book.fm Page 595 Monday, October 11, 2004 1:57 PM

Object Image Preprocessing

595

P [ eij′ ] =

eij′ β2

⋅e

−

( e ′ij )2 2 β2

,

(14.23)

and the transformation function is given by

eij′ = 2β 2 lg

1 . (1 − eij )

(14.24)

14.3 Detection of Boundary Edges The most stable features of compared object images are the boundary lines.5 This is characteristic of images of the same object across various ranges of the electromagnetic spectrum. There are many boundary edge detection algorithms. Methods of boundary edge detection are based, as a rule, on estimation of the gradient for each resolution element of the object image. Elements whose gradients exceed a certain brightness level are united as boundary lines with the same brightness characteristics, e.g., 0 or 1. Consider two boundary edge detection algorithms that can be applied in navigational systems constructed on the basis of the generalized approach to signal processing in the presence of noise. The ﬁrst algorithm is known as the Roberts operator or detection operator of boundary lines by four object image elements. This algorithm is based on the estimation and choice of image fragments with a high gradient level. If the digital object image is represented by the two-dimensional function g(i, j), then a gradient level at the point (i, j) is given by ∇g(i, j) ≈ R(i, j) =

[ g(i, j) − g(i + 1, j + 1)]2 + [ g(i, j + 1) − g(i + 1, j)]2 . (14.25)

Reference to Equation (14.24) shows that for image fragments with a constant brightness characteristic, the function R(i, j) is equal to zero. The function R(i, j) increases when the brightness characteristics of image elements change. The difference algorithm of absolute sum determination of diagonal gradients is more efﬁcient from a computational viewpoint:5 F(i, j) =|g(i, j) − g(i + 1, j + 1)|+|g(i, j + 1) − g(i + 1, j)|.

(14.26)

After gradient determination, we can carry out quantization according to

Copyright 2005 by CRC Press

1598_book.fm Page 596 Monday, October 11, 2004 1:57 PM

596

Signal and Image Processing in Navigational Systems  1, F(i, j) ≥ K g ,  Fq (i, j) =   0, F(i, j) < K g ,

(14.27)

where Kg is the threshold. The use of Equation (14.26) and Equation (14.27) allows us to form a boundary image in addition to the initial object image. For the second boundary edge detection algorithm, we take into account eight neighboring elements of the image element (i, j) for estimation of gradient.5 The gradient is determined by 3

(G, W )ij =

3

∑ ∑ g(i + K − 2, j + L − 2) ⋅ W(K , L) ,

(14.28)

K =1 L=1

where g(i, j) are the digital image elements; and W(K, L) is the weight function in matrix form (the matrix dimension is 3 × 3). The weight function has many forms:6 • The weight functions of a smooth gradient

W1 =

1

1

1

0

0 0

and

W2 =

-1 -1 -1

1

0

1

1

0 -1

1

0

1

1

0

-1

2

0

-2

1

0

-1

.

(14.29)

.

(14.30)

• The Soibel weight functions

W1 =

1

2

1

0

0

0

and

W2 =

-1 -2 -1 • The isotropic weight functions

W1 =

1

2

1

0

0

0

-1 - 2

-1

and

W2 =

The gradient level at the point (i, j) is equal to

Copyright 2005 by CRC Press

1

0

-1

2

0

- 2

1

0

-11

.

(14.31)

1598_book.fm Page 597 Monday, October 11, 2004 1:57 PM

Object Image Preprocessing

597

∇g(i, j)

= Sx2 (i, j) + Sy2 (i, j) ,

(14.32)

where Sx (i, j) = (G, W1 )

Sy (i, j) = (G, W2 ) .

and

(14.33)

By analogy with Equation (14.25) we can use the following formula: F(i, j) =|Sx (i, j)|+|Sy (i, j)|.

(14.34)

Quantization is carried out in accordance with Equation (14.26). The choice of the threshold Kg in Equation (14.27) has a great effect on the efﬁciency of the generalized signal and image processing algorithm. The threshold value depends on the object image characteristics: number of boundary lines, clearness of boundary lines, etc. There are three methods to automatically deﬁne the threshold.7 The ﬁrst method is called the method of the main boundaries. With this method, the threshold is chosen so that only the main boundary lines remain after digital image processing. The numbers of 1 and 0 are very high for an image; so the probability of false peaks of the correlation function is high, too. To circumvent these phenomena we use only the image elements with level 1, a technique that gives good results. The second method is called the method of averaging by image fragment. In this case, the gradient images obtained with Soibel weight functions are quantized on two levels based on the average value for nine neighboring image elements with the central point (i, j): i+1

A = 0.11

j+1

∑ ∑ F(m, n) .

(14.35)

m= i−1 n= j−1

The quantization equation takes the following form  1, Fq (i, j) =   0,

if F(i, j) ≥ kA,

(14.36)

otheerwise,

where k is the scale coefﬁcient. With a low value of k, the boundary lines of the image become thick and with high values of k, they become thin. An analogous approach is used in Reference 8.8 The third method is called the method of the mean and mean square deviation. In this method, there are two thresholds:

Copyright 2005 by CRC Press

1598_book.fm Page 598 Monday, October 11, 2004 1:57 PM

598

Signal and Image Processing in Navigational Systems L = m − Kσ

and

H = m + Kσ,

(14.37)

where k is the scale coefﬁcient (k ≥ 0). The quantization equation takes the following form:  1, Fq (i, j) =   0,

F(i, j) ≥ H , F(i, j) ≤ L.

(14.38)

Thus, the image elements with gradient levels belonging to the interval [H, L] are not taken into consideration with the use of the generalized image processing algorithm. Rejection of image elements with gradient levels that are close to the mean reduces noise.

14.4 Conclusions The discussion in this chapter allows us to draw the following conclusions. There are two types of object image distortions in form: the geometrical distortions and the distortions in brightness characteristics. Sources of these distortions are dissimilar. A source of the geometrical distortions is inaccurate information regarding the object’s position in space. As a rule, the geometrical distortions can appear during transmission of object images by various sensors when an implementation of these sensors assumes various tracking angles. Sources of the distortions in brightness characteristics of object images are changes in meteorological or seasonal conditions. If there are various types of sensors, for example, the radar side-scanning sensors and the optical scanning sensors, it is necessary to carry out one-toone correspondence with a geometrical transformation or correction. Two techniques can be applied with the use of the generalized image processing algorithm based on the generalized approach to signal processing in the presence of noise. These are the perspective transform and the polynomial estimation. With the use of the ﬁrst technique, it is necessary to ﬁrst transform the object image in the radar-side-scanning plane to the coordinate system of angles θ, γ, and ψ. The second technique is widely used because additional information is not required and the information included in the obtained object images is sufﬁcient. If the same objects are obtained by various sensors, the distributions of brightness characteristics are different. It is necessary to carry out one-toone correspondence with the object images. For this purpose, two transformation techniques can be used. The ﬁrst technique is based on knowledge of statistical characteristics of object images. The second technique does not require that information because the radar object image is transformed so Copyright 2005 by CRC Press

1598_book.fm Page 599 Monday, October 11, 2004 1:57 PM

Object Image Preprocessing

599

that the brightness characteristics correspond to the exponential distribution law. The boundary lines are the most stable features of compared object images. The boundary lines can be considered as a characteristic of images of electromagnetic spectrum. Many boundary edge detection algorithms are constructed. The main principle of boundary edge detection is based on estimation of the gradient for each resolution element of the object image.

References 1. Wong, R., Sensor transformations, IEEE Trans., Vol. SMC-7, No. 12, 1977, pp. 836–841. 2. Lipkin, B. and Rosenfeld, A., Eds., Picture Processing and Psychopictories, New York, 1970. 3. Berry, J. and Yoo, J., Geometric preprocessing of sensor data used for image matching, in Proceedings of the SPIE, Vol. 186, 1979, pp. 2–11. 4. Wong, R. and Hall, E., Image transformations, in Proceedings of the IEEE 4th International Joint Conference on Pattern Recognition, 1978, pp. 939–942. 5. Boland, J., Peters, E. et al., Automatic correlation of non-compatible imaging systems, in Proceedings of IEEE SOUTHEASTCON, 1979, pp. 230–233. 6. Frei, W. and Chen, C., Fast boundary detection: a generalization and a new algorithm, IEEE Trans., Vol. C-26, No. 10, 1977, pp. 988–998. 7. Boland, J. et al., A pattern recognition technique for scene matching of dissimilar imagery, in Proceedings of the IEEE 18th Conference on Decision and Control, 1979, pp. 806–811. 8. Robinson, G., Detection and coding of edges using directional masks, Opt. Eng., Vol. 16, No. 6, 1977, pp. 653–667.

Copyright 2005 by CRC Press

1598_A01.fm Page 601 Friday, October 1, 2004 10:53 AM

Appendix I Classification of Stochastic Processes

In analysis of the considered problems, we investigated many types of stationary and nonstationary stochastic processes (the target return signal). So it is necessary to classify the analyzed stochastic processes. The basis of this classification is the behavior of the two-dimensional correlation function R(t, τ) of fluctuations of the target return signal, where τ = t2 – t1 and t = 0.5(t1 + t2). We take into consideration only those general features of the behavior of R(t, τ) that significantly influence the structure of the two-dimensional power spectral density S(ω, Ω) of target return signal fluctuations, namely, the periodic or nonperiodic dependence of R(t, τ) on the variable τ and on the variable t in the case of the nonstationary target return signal. The function of τ defines stochastic (random) features of the target return signal and the function of t defines regular (deterministic) features of the target return signal. From the viewpoint of regular features of the target return signal, we can define stationary processes (see Figure I.1–Figure I.3) and two forms of nonstationary processes: nonperiodic (Figure I.4–Figure I.6) and periodic (Figure I.7–Figure I.9) target return signals, depending on the time t. From the viewpoint of stochastic processes, we can define nonperiodic target return signals (Figure I.1, Figure I.4, and Figure I.7), periodic target return signals (Figure I.2, Figure I.5, and Figure I.8), and the quasiperiodic target return signals (Figure I.3, Figure I.6, and Figure I.9), in which the correlation function R(t, τ) can be both periodic and nonperiodic with respect to the variable τ. The simplest example of this correlation function is the product discussed in Chapter 3, Chapter 6, and Chapter 9: R(τ) = R1(τ) × R2(τ – nTp). In addition, we discussed more complex correlation function (see Chapter 4 and Chapter 10). We now list some problems where we used the correlation functions and power spectral densities of target return signal fluctuations shown in Figure I.1–Figure I.9: 1. Doppler fluctuations of the target return signal with the simple harmonic searching signal; chaotic motion of scatterers with the simple harmonic searching signal; fluctuations of the transformed target return signal with the linear frequency-modulated searching signal in the case of stationary radar (see Figure I.1); line or conical Copyright 2005 by CRC Press

1598_A01.fm Page 602 Friday, October 1, 2004 10:53 AM

602

Signal and Image Processing in Navigational Systems

scanning with the simple harmonic searching signal if the radar is stationary (Figure I.2); spiral scanning with the simple harmonic searching signal if the radar is stationary; line scanning with chaotic motion of scatterers and the simple harmonic searching signal if the radar is stationary; rotation of the radar antenna polarization plane and chaotic motion of scatterers with the simple harmonic searching signal if the radar is stationary (Figure I.3). 2. Doppler fluctuations of the target return signal under the continuous searching signal with nonperiodic variations in power or the radar moves with acceleration (Figure I.4); line or conical scanning under the continuous searching signal with nonperiodic variations in power if the radar is stationary (Figure I.5); spiral scanning under the simple harmonic searching signal with nonperiodic variations in power if the radar is stationary; line scanning under chaotic motion of scatterers and the simple harmonic searching signal with nonperiodic variations in power if the radar is stationary; rotation of the radar antenna polarization plane and chaotic motion of scatterers with nonperiodic variations in power under the simple harmonic searching signal if the radar is stationary (Figure I.6). 3. Doppler fluctuations of the target return signal under the continuous searching signal with periodic variations in power or in the velocity of moving radar (Figure I.7); target return signal fluctuations under scanning of the three-dimensional (space) target with the pulsed searching signal if the radar is stationary; segment scanning with the simple harmonic searching signal if the radar is stationary (Figure I.8); target return signal fluctuations under scanning of the threedimensional (space) target by the pulsed searching signal if the radar is moving; line scanning with the simple harmonic searching signal if the radar is moving (Figure I.9). The main reason for the nonstationary state of the target return signal in these problems is the dependence of the amplitude variance (or power) of the target return signal on time (the constant component in all considered cases is equal to zero because we consider the radio signal). Using the main results discussed in Chapter 8–Chapter 10, we can obtain the correlation functions and power spectral densities of target return signal fluctuations, in which the frequency of the searching signal is a function of time. The proposed classification of stochastic processes given here (the target return signal) is true, naturally, only within the limits of correlation theory and does not exclude other forms of classification.1,2

Copyright 2005 by CRC Press

1598_A01.fm Page 603 Friday, October 1, 2004 10:53 AM

Classification of Stochastic Processes

603 t

R(t, τ)

τ (a)

S (ω, t )

t

ω (b) S(ω, Ω)

Ω

ω (c) FIGURE I.1 (a) The instantaneous correlation function R(t, τ), (b) instantaneous power spectral density S(ω, t), and (c) two-dimensional power spectral density S(ω, Ω) of stationary stochastic processes — nonperiodic stochastic processes.

Copyright 2005 by CRC Press

1598_A01.fm Page 604 Friday, October 1, 2004 10:53 AM

604

Signal and Image Processing in Navigational Systems t

R(t, τ)

τ (a) t

S(ω, t )

ω (b) S(ω, Ω)

Ω

ω (c) FIGURE I.2 (a) The instantaneous correlation function R(t, τ), (b) instantaneous power spectral density S(ω, t), and (c) two-dimensional power spectral density S(ω, Ω) of stationary stochastic processes — periodic stochastic processes.

Copyright 2005 by CRC Press

1598_A01.fm Page 605 Friday, October 1, 2004 10:53 AM

Classification of Stochastic Processes

605 t

R(t, τ)

τ (a) t

S(ω, t)

ω (b) S (ω, Ω)

Ω

ω (c) FIGURE I.3 (a) The instantaneous correlation function R(t, τ), (b) instantaneous power spectral density S(ω, t), and (c) two-dimensional power spectral density S(ω, Ω) of stationary stochastic processes — quasiperiodic stochastic processes.

Copyright 2005 by CRC Press

1598_A01.fm Page 606 Friday, October 1, 2004 10:53 AM

606

Signal and Image Processing in Navigational Systems t

R(t, τ)

τ (a) t

S(ω, t)

ω (b) S (ω, Ω)

Ω

ω

(c) FIGURE I.4 (a) The instantaneous correlation function R(t, τ), (b) instantaneous power spectral density S(ω, t), and (c) two-dimensional power spectral density S(ω, Ω) of nonstationary nonperiodic processes — nonperiodic stochastic processes.

Copyright 2005 by CRC Press

1598_A01.fm Page 607 Friday, October 1, 2004 10:53 AM

Classification of Stochastic Processes

607

R(t, τ)

t

τ (a) t

S(ω, t)

ω (b) S (ω, Ω)

Ω

ω

(c) FIGURE I.5 (a) The instantaneous correlation function R(t, τ), (b) instantaneous power spectral density S(ω, t), and (c) two-dimensional power spectral density S(ω, Ω) of nonstationary nonperiodic processes — periodic stochastic processes.

Copyright 2005 by CRC Press

1598_A01.fm Page 608 Friday, October 1, 2004 10:53 AM

608

Signal and Image Processing in Navigational Systems R(t, τ)

t

τ (a) t

S(ω, t)

ω (b) S (ω, Ω)

Ω

ω

(c) FIGURE I.6 (a) The instantaneous correlation function R(t, τ), (b) instantaneous power spectral density S(ω, t), and (c) two-dimensional power spectral density S(ω, Ω) of nonstationary nonperiodic processes — quasiperiodic stochastic processes.

Copyright 2005 by CRC Press

1598_A01.fm Page 609 Friday, October 1, 2004 10:53 AM

Classification of Stochastic Processes

609

R(t, τ)

t

τ (a)

S(ω, t)

t

ω (b) S (ω, Ω)

Ω

ω

(c) FIGURE I.7 (a) The instantaneous correlation function R(t, τ), (b) instantaneous power spectral density S(ω,t), and (c) two-dimensional power spectral density S(ω, Ω) of nonstationary periodic processes — nonperiodic stochastic processes.

Copyright 2005 by CRC Press

1598_A01.fm Page 610 Friday, October 1, 2004 10:53 AM

610

Signal and Image Processing in Navigational Systems R(t, τ)

t

τ (a)

S(ω, t)

t

ω (b) S (ω, Ω)

Ω

ω

(c) FIGURE I.8 (a) The instantaneous correlation function R(t, τ), (b) instantaneous power spectral density S(ω, t), and (c) two-dimensional power spectral density S(ω, Ω) of nonstationary periodic processes — periodic stochastic processes.

Copyright 2005 by CRC Press

1598_A01.fm Page 611 Friday, October 1, 2004 10:53 AM

Classification of Stochastic Processes

611

R(t, τ)

t

τ (a)

S(ω, t)

t

ω (b) S (ω, Ω)

Ω

ω

(c) FIGURE I.9 (a) The instantaneous correlation function R(t, τ), (b) instantaneous power spectral density S(ω, t), and (c) two-dimensional power spectral density S(ω, Ω) of nonstationary periodic processes — quasiperiodic stochastic processes.

Copyright 2005 by CRC Press

1598_A01.fm Page 612 Friday, October 1, 2004 10:53 AM

612

Signal and Image Processing in Navigational Systems

References 1. Tikhonov, V., Statistical Radio Engineering, Radio and Svyaz, Moscow, 1982 (in Russian). 2. Romanenko, A. and Sergeev, G., Problems of Applied Analysis of Stochastic Processes, Soviet Radio, Moscow, 1968 (in Russian).

Copyright 2005 by CRC Press

1598_book.fm Page 613 Monday, October 11, 2004 1:57 PM

Appendix II The Power Spectral Density of the Target Return Signal with Arbitrary Velocity Vector Direction of Moving Radar in Space and with the Presence of Roll and Pitch Angles

The main parameters of the target return signal from the three-dimensional (space) target — the power, the instantaneous and average frequency, and the power spectral density bandwidth of target return signal ﬂuctuations — depend on the radar velocity and position of the radar antenna directional diagram with respect to the velocity vector of moving radar and the reﬂecting surface. Before, we believed that the position of the directional diagram was deﬁned by angles reckoned from the velocity vector of moving radar. In fact, the direction of the velocity vector of moving radar is deﬁned relative to the axes of the immovable coordinate system of the Earth, and the position of the directional diagram is deﬁned with respect to the axes of aircraft. Due to vibrations of aircraft, the directional diagram can change its position with respect to the velocity vector of moving radar and the underlying reﬂecting surface that leads to changes in all characteristics and parameters of the power spectral density of target return signal ﬂuctuations. Introduce three Cartesian coordinate systems: the immovable horizontal coordinate system OXYZ (see Figure II.1), whose axes are parallel to the axes of the immovable coordinate system of the Earth and whose center O is at the aircraft’s center of gravity; the coordinate system OX1Y1Z1 localized by the aircraft, whose center coincides with the center of the horizontal coordinate system OXYZ (see Figure II.1), and whose axes, in the general case, are rotated by the angles of heading ψc , pitch θc , and roll γc with respect to the horizontal coordinate system OXYZ. Because we assume that the radar antenna is hardly connected with the aircraft, ψc = 0, and under the condition ψc = θc = γc = 0°, the coordinate system OXYZ coincides with the coordinate system of the aircraft OX1Y1Z1; the intermediate coordinate system OX′Y′Z′ is shown in Figure II.1 to illustrate a jump from the coordinate system OXYZ to the coordinate system OX1Y1Z1. In the general case, a vector deﬁning the Doppler shift in the target return signal frequency is the sum of three vectors

Copyright 2005 by CRC Press

1598_book.fm Page 614 Monday, October 11, 2004 1:57 PM

614

Signal and Image Processing in Navigational Systems Y

Y′ Y1

X1

θc

γc

X′

θc ψc

X

O

ψc

γc

Z Z′

Z1

FIGURE II.1 The coordinate systems OXYZ and OX1Y1Z1.

VΣ = V − U + [r , ω ] ,

(II.1)

where V is the velocity vector of the moving aircraft’s center of gravity relative to the immovable reﬂecting surface; U is the velocity vector of moving scatterers of the underlying surface, for example, a sea surface; [r, ω] is the vector of linear velocity of the moving radar antenna caused by angle displacement of the aircraft relative to its center of gravity when the radar antenna is spaced at a distance r from the center of gravity of the aircraft, and ω is the vector of angular velocity of the aircraft. Based on Equation (II.1) we can express the vector V in the horizontal coordinate system in the following form (see Figure II.2): V = { Vx , Vy , Vz } = {V cos ε 0 cos α , V sin ε 0 , − V cos ε 0 sin α} .

(II.2)

The direction “radar–locality” deﬁned by the basis vector I1 can be represented in the coordinate system OX1Y1Z1 in the following form (see Figure II.3): I1 = { lx , ly , lz } = {cos γ 1 cos β1 , − sin γ 1 , − cos γ 1 sin β1 } ,

Copyright 2005 by CRC Press

(II.3)

1598_book.fm Page 615 Monday, October 11, 2004 1:57 PM

Power Spectral Density of the Target Return Signal

615

Y V −Vy Vz

ε0

α O

X Vx

Z FIGURE II.2 The velocity vector V of moving radar in the coordinate system OXYZ. Y1 V X1 −β1 θ γ1 Z1

I X

−β1

FIGURE II.3 The basis vector I in the coordinate system OX1Y1Z1.

Copyright 2005 by CRC Press

1598_book.fm Page 616 Monday, October 11, 2004 1:57 PM

616

Signal and Image Processing in Navigational Systems

where ε0 is the trajectory angle of the aircraft and α is the drift angle of the aircraft. The angle between the vector V and the basis vector Ih has the following form: cos(V ⋅ Ih ) = cos θ = cos ε 0 cos α cos γ h cos β h – sin ε 0 sin γ h − cos ε 0 sin α cos γ h sin β h

,

(II.4)

= cos ε 0 cos γ h cos(α + β h ) − sin ε 0 sin γ h where cos γ h cos β h = cos γ 1 cos β1 cos θc + sin γ 1 sin θc cos γ c − cos γ 1 sin β1 sin γ c sin θc sin γ h = sin γ 1 cos θc cos γ c − cos γ 1 cos β1 sin θc

;

(II.5)

;

(II.6)

cos γ h sin β h = cos γ 1 sin β1 cos γ c + sin γ 1 sin γ c .

(II.7)

− cos γ 1 sin β1 sin γ c cos θc

The angles γh and βh deﬁne the position of the basis vector Ih in the coordinate system OXYZ, as with the basis vector I1. To deﬁne the correlation function of target return signal ﬂuctuations, it is necessary to substitute the formula ∆ρ = – V · τ cosθ in the general formula using the cosθ-expansion in terms of Equation (II.5)–Equation (II.7) with respect to the differentials ϕ and ψ in the coordinate system OX1Y1Z1. As the directional diagram is given in the coordinate system OX1Y1Z1, it does not vary. Based on Equation (II.6) we must deﬁne the angle γh , for which it is necessary to estimate the function S°(γh) using experimental or theoretical data. Because the angle γh0 corresponds to the direction of the directional diagram in the coordinate system OXYZ, the function can be estimated in the following form: 2

S°( γ h ) = S°( γ h0 ) ⋅ e k1ψ h + k2ψ h ,

(II.8)

where ψh = γh – γh0. This approximation is sufﬁciently close when the directional diagram is not very wide (by 20–30˚). Also, it is only necessary to take the factor k2 into consideration near the value γh ≈ 90° in the case of a smooth sea surface. As well known, the value of k1 in the case of a ground surface is not very high, but in the case of a smooth sea surface, the value of k1 reaches 20–30 at γh = 60–80° and the value of k2 reaches approximately 200.1

Copyright 2005 by CRC Press

1598_book.fm Page 617 Monday, October 11, 2004 1:57 PM

Power Spectral Density of the Target Return Signal

617

In the case of slope scanning, the parameter k1(γh) can be deﬁned by experimental data k1 ( γ h0 ) = 13.2 ⋅

10 lg S°(γ h0 −

∆v 2 2

) − 10 lg S°(γ h

0

+

∆v 2 2

)

∆v

,

(II.9)

2

where the value of ∆v is given in degrees. We can deﬁne other parameters depending on the angle γh, for example, Td =

2h . c sin γ h

(II.10)

For this purpose, we have to express the function of the angle γh in the form of a series expansion with respect to the variables ϕ and ψ to deﬁne a function relating the angle ψh to the variables ϕ and ψ using Equation (II.6). For example, deﬁne the power spectral density of target return signal ﬂuctuations with the nonmodulated searching signal in the case of slope scanning. We assume that the directional diagram g(ϕ, ψ) can be approximated by the Gaussian law and that the function S°(γh) can be approximated by the exponent under the condition k2 = 0. We can show that the power spectral density is Gaussian:2 −π p S(ω ) = ⋅e ∆Ω

(ω − Ω )2 ∆Ω2

;

(II.11)

Ω = ω 0 + Ω0 + ∆Ω k + ∆Ωρ ; p=

PS λ 2G02 S°( γ h0 )∆ h ∆ v sin γ h0 128π 3 h2

⋅e

0.125 π ( kh2 ∆(h2 ) + kv2 ∆(v2 ) ) 1 1

(II.12)

;

∆Ω = 2 2 πλ −1 ∆ (v2 )V12 + ∆ (h2 )V22 ; V1 = − (Vx1 sin γ 0 cos β 0 + Vy1 cos γ 0 + Vz1 sin γ 0 sin β 0 ) and V2 = − (Vx1 sin β 0 − Vz1 cos β 0 ) ;  Vx1 = Vx cos θc ;    Vy1 = − Vx sin θc cos γ c + Vy cos γ c cos θc + Vz sin γ c ;   Vz1 = Vx sin γ c sin θ c −Vy sin γ c cos θc + Vz cos γ c ;

Copyright 2005 by CRC Press

(II.13)

(II.14)

(II.15)

(II.16)

1598_book.fm Page 618 Monday, October 11, 2004 1:57 PM

618

Signal and Image Processing in Navigational Systems

Here, the angles γ0 and β0 deﬁne the direction of the directional diagram axis in the coordinate system OX1Y1Z1; Vx1, Vy1, and Vz1 are the components of the vector V in the coordinate system OX1Y1Z1; Ω0 = 4πV λ −1 cos θ0 = ΩDx1 + ΩDy 1 + ΩDz1 ; ΩDx 1 = 4πVx1 λ −1 cos γ 0 cos β 0 ;

(II.17)

ΩDy 1 = − 4πVy1 λ −1 sin γ 0 ; and ΩDz1 = − 4πVz1 λ −1 cos γ 0 sin β 0 ; ∆Ω k = δ x1 ΩDx 1 + δ y1 ∆ΩDy 1 + δ z1 ∆ΩDz1

and (II.18)

∆Ωρ = δ ρx ΩDx 1 + δ ρy ΩDy 1 + δ ρz ΩDz1 ; δ x1 = − δ y1 =

k1 ∆ (v2 ) ctg γ 0 ; and 4π

δ z1 = −

δ ρx =

kv ∆ (v2 ) sin γ 0 cos β 0 + kh ∆ (h2 ) sin β 0 ; 4π cos γ 0 co os β 0

kv ∆ (v2 ) sin γ 0 sin β 0 + kh ∆ (h2 ) cos β 0 ; 4π cos γ 0 sin β 0

δ x1 ctg γ h0 k1

(II.19)

; δ ρy =

δ y1 ctg γ h0 k1

; δ ρz =

δ z1 ctg γ h0 k1

;

(II.20)

 cos γ 0 cos θc cos γ c + sin γ 0 cos β 0 sin θc + sin γ 0 sin β 0 sin γ c cos θc ;  α1 = sin γ h0   sin β 0 sin θc − cos β 0 sin γ c cos θc  ;  α2 = sin γ h0  (II.21) kv = α 1 k1 tg γ h0 ; kv1 = α 1 ( k1 + ctg γ h0 )tg γ h0 ; kh = α 2 k1 tg γ h0 ; and

(II.22)

kh1 = α 2 ( k1 + ctg γ h0 )tg γ h0 , where p is the target return signal power; and PS is the power of the searching signal [see Equation (II.13)]. It should be pointed out that in the majority of Copyright 2005 by CRC Press

1598_book.fm Page 619 Monday, October 11, 2004 1:57 PM

Power Spectral Density of the Target Return Signal

619

cases, with low values of the parameters ∆h and ∆v , we can limit ourselves to the ﬁrst factor and there is no need to take the second factor into consideration to deﬁne the target return signal power. For example, assuming that ∆h = ∆v = 6°, γc = θc = 0°, γ0 = 65°, k1 = 20, we can write

e

( k1 + ctg γ 0 )2 ∆(v2 ) 8π

= 1.172

or 0.7 dB .

(II.23)

The effective bandwidth ∆Ω given by Equation (II.14) depends on the directional diagram width in both the planes ∆h and ∆v . The parameter ∆Ωk given by Equation (II.18)–Equation (II.22) is the shift in the average frequency of the power spectral density of target return signal ﬂuctuations caused by the effect of the function S°(γ). The parameter ∆Ωρ given by Equation (II.18) — Equation (II.22) is the shift in the average frequency of the power spectral density of target return signal ﬂuctuations caused by distance variations within the directional diagram. The parameter ∆Ωρ is k1 times less than the parameter ∆Ωk [see Equation (II.18) — Equation (II.22)]. Assume that roll and pitch are absent, i.e., that the condition γc = θc = 0° is satisﬁed. In this case, the components of the velocity vector of moving radar for both coordinate systems OXYZ and OX1Y1Z1 are the same: γ h0 = γ 0 , α 1 = ctg γ 0 , α 2 = 0, kh = 0, kv = k1 ; δx = δz = − δy =

k1 ∆ (v2 ) tg γ 0 ; 4π

(II.24)

k1 ∆ (v2 ) ctg γ 0 ; 4π

δ ρx = δ ρ z = − δ ρy =

∆ (v2 ) ; and 4π

∆ (v2 ) ctg 2 γ 0 ; 4π

∆Ω = 2 2 πλ −1 ∆ (v2 )V12 + ∆ (h2 )V22 ; V1 = − V[cos ε 0 sin γ 0 cos(α + β 0 ) + sin ε 0 cos γ 0 ] = Va1 V2 = − V cos ε 0 sin(α + β 0 ) = Vb1 .

Copyright 2005 by CRC Press

(II.25) and

(II.26)

1598_book.fm Page 620 Monday, October 11, 2004 1:57 PM

620

Signal and Image Processing in Navigational Systems

It should be noted that the relative shifts δx , δz , δρx , and δρz of the horizontal components ΩDx and ΩDz are the same and increase with an increase in the angle γ0 [see Equation (II.24)]. The shifts δy and δρy have a different sign because the effects of the function So(γ) and distance ρ increase the effective angle γ0 , and the shifts δy and δρy increase with a decrease in the angle γ0. An increase in the angle γ0 leads to a decrease in the view of the vector V on the X and Z axes and to an increase in the view of the vector V on the Y axis. The values of the shifts δρx and δρz are constant for a given directional diagram and independent of the angle γ0 , but the shift δρy depends mainly on the angle γ0. We deﬁne the effect of each component of the velocity vector of moving radar on the effective power spectral density bandwidth of target return signal ﬂuctuations. If only one component of the velocity vector of moving radar is different from zero, we can write ∆Ω x = 2 2 πVx λ −1 ∆ (v2 ) sin 2 γ 0 cos 2 β 0 + ∆ (h2 ) sin 2 β 0 ; ∆Ω y = 2 2 πVy λ −1 ∆ v cos γ 0 ; and

(II.27)

∆Ω z = 2 2 πVz λ −1 ∆ (v2 ) sin 2 γ 0 sin 2 β 0 + ∆ (h2 ) cos 2 β 0 . It should be pointed out that in the case of the directional diagram with axial symmetry, there is no need to carry out complex mathematics for aircraft vibrations. It is necessary to deﬁne only the angles βh0 and γh0 using Equation (II.5)–Equation (II.7) in the coordinate system OXYZ that corresponds to the direction of the directional diagram axis and to deﬁne the parameter k1(γh0) using the function S°(γ) for the determined angle γ0. All characteristics of the power spectral density of target return signal ﬂuctuations are deﬁned for these values of the angles under the condition γc = θc = 0°. This technique of deﬁning the power spectral density S(ω) of target return signal ﬂuctuations (with a nonhorizontally moving radar and with the angles of roll γc and pitch θc) can be used for all types of searching signals.

References 1. Kolchinsky, V., Mandurovsky, I., and Konstantinovsky, M., Doppler Devices and Navigational Systems, Soviet Radio, Moscow, 1975 (in Russian). 2. Winitzky, A., Basis of Radar under Continuous Generation of Radio Waves, Soviet Radio, Moscow, 1961 (in Russian).

Copyright 2005 by CRC Press

Signal and Image Processing in Navigational Systems (Electrical Engineering & Applied Signal Processing Series)

Signal and Image Processing in Navigational Systems

Medical Image Analysis Methods (Electrical Engineering & Applied Signal Processing)

Applications in Time-Frequency Signal Processing (Electrical Engineering & Applied Signal Processing Series)

Advanced Signal Processing and Noise Reduction, 2nd Edition (Electrical Engineering & Applied Signal Processing Series)

Smart Antennas (Electrical Engineering & Applied Signal Processing Series)

Smart Antennas (Electrical Engineering & Applied Signal Processing Series)

Handbook of Neural Network Signal Processing (Electrical Engineering & Applied Signal Processing Series)

Pattern Recognition in Speech and Language Processing (Electrical Engineering & Applied Signal Processing Series)

VLSI Signal Processing Systems

Signal Processing and Linear Systems

Signal Processing

Signal Processing and Linear Systems

Signal Processing and Linear Systems

Signal Processing

Signal Processing and Linear Systems

Optimisation in Signal and Image Processing

Advances in Nonlinear Signal and Image Processing (EURASIP Book Series on Signal Processing and Communications)

Multidimensional Digital Signal Processing (Prentice-Hall Signal Processing Series)

Wireless Communications: Signal Processing Perspectives (Prentice Hall Signal Processing Series)

Handbook of Signal Processing Systems

Signal Processing and Multimedia

Electrical Measurement, Signal Processing, and Displays

Evolutionary image analysis and signal processing

Adaptive Blind Signal and Image Processing

Evolutionary Image Analysis and Signal Processing

Two dimentional signal and image processing

Digital signal and image processing using MATLAB

Adaptive Blind Signal and Image Processing

Signal and Image Processing for Remote Sensing

Noise Reduction in Speech Applications (Electrical Engineering & Applied Signal Processing Series)

Signal and Image Processing in Navigational Systems (Electrical Engineering & Applied Signal Processing Series)

Signal and Image Processing in Navigational Systems

Medical Image Analysis Methods (Electrical Engineering & Applied Signal Processing)

Applications in Time-Frequency Signal Processing (Electrical Engineering & Applied Signal Processing Series)

Advanced Signal Processing and Noise Reduction, 2nd Edition (Electrical Engineering & Applied Signal Processing Series)

Smart Antennas (Electrical Engineering & Applied Signal Processing Series)

Smart Antennas (Electrical Engineering & Applied Signal Processing Series)

Handbook of Neural Network Signal Processing (Electrical Engineering & Applied Signal Processing Series)

Pattern Recognition in Speech and Language Processing (Electrical Engineering & Applied Signal Processing Series)

VLSI Signal Processing Systems

Signal Processing and Linear Systems

Signal Processing

Signal Processing and Linear Systems

Signal Processing and Linear Systems

Signal Processing

Signal Processing and Linear Systems

Optimisation in Signal and Image Processing

Advances in Nonlinear Signal and Image Processing (EURASIP Book Series on Signal Processing and Communications)

Multidimensional Digital Signal Processing (Prentice-Hall Signal Processing Series)

Wireless Communications: Signal Processing Perspectives (Prentice Hall Signal Processing Series)

Handbook of Signal Processing Systems

Signal Processing and Multimedia

Electrical Measurement, Signal Processing, and Displays

Evolutionary image analysis and signal processing

Adaptive Blind Signal and Image Processing

Evolutionary Image Analysis and Signal Processing

Two dimentional signal and image processing

Digital signal and image processing using MATLAB

Adaptive Blind Signal and Image Processing

Signal and Image Processing for Remote Sensing

Noise Reduction in Speech Applications (Electrical Engineering & Applied Signal Processing Series)

Recommend Documents